Photogrammetric

Applications for
Cultural Heritage
Guidance for Good Practice
Summary

This guidance covers the practical application of photogrammetry in recording cultural

heritage, with particular reference to structure from motion (SfM) techniques. Our

audience for this document includes survey contractors, archaeological contractors,

voluntary organisations and specialists. Photogrammetric image acquisition and

processing, until recently requiring a considerable investment in hardware and

software, are now possible at a fraction of their former cost. This has led to a huge

increase in the use of photogrammetry in cultural heritage recording. The skills

required to apply the techniques successfully and accurately are discussed, and

background information on how various parts of the process work is provided so that

better results can be achieved through better understanding.

Photogrammetry is characterised by its versatility, and is applicable over a wide range

of scales, from landscapes to small objects. The particular requirements needed at

these different scales are outlined, and both imaging techniques and useful ancillary

equipment are described. The different types of outputs are discussed, including their

suitability for further interrogation using a range of established analytical techniques

and the presentation options available. A range of case studies illustrates the

application of photogrammetry across a variety of projects that broadly reflect

the areas discussed in the text. This document is one of a number of Historic England

technical advice documents on how to survey historic places.

This document has been prepared by Jon Bedford. This edition published by

Historic England October 2017. All images © Historic England unless

otherwise stated.

Please refer to this document as:

Historic England 2017 Photogrammetric Applications for Cultural Heritage. Guidance for

Good Practice. Swindon. Historic England.

HistoricEngland.org.uk/advice/technical-advice/recording-heritage/

Front cover
13th century roof boss depicting Samson wrestling
a lion, Hailes Abbey, Gloucestershire.
Contents

Introduction.........................................1
3.4 Terrestrial photogrammetry

for smaller objects .....................................73

Core skills .................................................................1

3.5 Textured 3D models ...................................77

Appropriate use of different

3.6 3D printing..................................................78

survey techniques....................................................2
3.7 Community involvement ...........................79

3.8 Metadata and archive ................................79

1 Basic Principles ...........................4

4 Case Studies ..............................80

1.1 Basic procedure ...........................................4

1.2 The chief ray and the principles

Case Study 1: Using archive aerial imagery ........80

of intersection ..............................................6
Case Study 2: SUA imagery on single sites..........83

1.3 Collinearity ...................................................7

Case Study 3: SUA landscape survey. ..................87

1.4 Interior and exterior orientation.................7

Case Study 4: Combining SUA and

1.5 The SfM–MVS photogrammetric process ...8

terrestrial data .......................................................90

1.6 Measurement from single, paired

Case Study 5: Terrestrial imaging on

or multiple images.......................................9
archaeological sites...............................................95

1.7 Sources of error..........................................13

Case Study 6: Underwater

1.8 General workflow .......................................14

photogrammetric survey.....................................102

Case Study 7: Modelling small objects ..............107

2 General Considerations .............19

5 References...............................110

2.1 Capturing the initial data ..........................19

2.2 Image arrangement ...................................34

Software ...............................................................113

2.3 Control ........................................................42

Other web resources............................................114

2.4 Historical imagery......................................50

6 Glossary ..................................115

3 Scales and Applications .............53

3.1 Aerial photogrammetry for archaeology,

Acknowledgements..........................122

landscapes and buildings .........................53

3.2 Terrestrial photogrammetry for buildings

Contact Historic England ....................................123

and structures ............................................66

3.3 Terrestrial photogrammetry

for excavations ...........................................73

Introduction

This guidance covers the application of photogrammetry in cultural heritage with

particular reference to structure from motion (SfM) techniques. The aim is to raise
awareness of the issues that are commonly encountered, rather than to deal with all
aspects of photogrammetric processing in depth. There are many potential uses for
this technique across a wide variety of scales, ranging from landscapes (for example
10km2 or more) to small objects (for example 10mm3). The different outputs are
suitable for a range of established analytical techniques and offer many
presentation options.

Photogrammetric image acquisition and Photogrammetry

processing, until recently requiring a considerable You need an understanding of the image
investment in hardware, software, expertise arrangements needed to achieve the best
and time, is now possible at a fraction of its coverage of the target and ensure the highest
former cost. The advent of more affordable accuracy of the models produced. You will need
photogrammetric software over the last few years, some knowledge of the camera distortions that
in particular the use of SfM techniques, has seen can affect the quality of the results.
an explosion in its use in archaeology. This has
been helped by the development of relatively Survey
cheap digital cameras capable of capturing You need an understanding of the appropriate
high-quality imagery, and by advances in the level of detail, scale, orientation and control
processing capability of personal computers. for different projects. You need to know when
Additionally, the development of small unmanned it is appropriate to use photogrammetry and
aircraft (SUA), which can be used to capture low- when another surveying method would be more
level aerial imagery, has contributed by providing suitable to achieve the desired product.
an easy method for recording individual sites.
You need to be able to organise the stages of
a project in an efficient way. You also need
Core skills an understanding of the importance of metric
accuracy, and of the methods available to
In order to use photogrammetric techniques quantify and improve it.
successfully in archaeology, a number of core
skills are required. Software
You not only need to know about
Photography photogrammetric software packages, but also
You need an understanding of how to use cameras those required for image processing and later
and related equipment to achieve the best digitisation and analysis, such as computer-aided
possible images in different circumstances. design (CAD) software and geographic information
systems (GIS).

< < Contents

Archaeology/architecture (TSTs) and hand-drawn survey, are usually
You need to be able to interpret and use the data characterised by the selection of data measured
generated to answer the questions being asked. at the point of capture (Figure 1).

Data presentation One survey technique on its own rarely provides

You need to be able to present the data in a clear, the perfect solution for most recording and
unambiguous and aesthetically pleasing format, analysis projects, and often requires augmenting
using established conventions. with data derived from different sources or
methods. Your focus should be on the suitability
These skills can be shared across a team, but it is of the products for the task at hand and the
important that all are present. required deliverables rather than the method used
to derive them. Similarly, although the accuracy
of photogrammetric reconstruction is affected by
Appropriate use of different a number of factors, it is wise to remain focused
survey techniques on the desired performance and use of the
product rather than obtaining the highest possible
Broadly speaking, survey techniques can be accuracy and density of reconstruction under all
divided into mass capture and selective methods. circumstances. Photogrammetrically processed
Mass capture methods, such as three-dimensional imagery can produce resolutions far in excess
(3D) laser scanning and photogrammetry, of the required result, which often leads to time
are characterised by large amounts of being spent on processing unnecessary data,
undifferentiated data at the point of capture the handling of very large files and subsequently
and selection from that data off-site. Selective the need to decimate the derived data in order
methods, such as global navigation satellite to achieve a usable and suitable product. For
systems (GNSS), use of total station theodolites example, specifying a 10mm ground sample

Figure 1
Hillshaded DEM derived from mass capture techniques
(left); highly selective interpretative linework derived
by GNSS survey of the same area (right).

< < Contents 2

distance (GSD; the distance on the ground (DSLRs) are almost always better than those
between adjacent pixel centres in the image) for from point-and-shoot cameras, the problems
vertical aerial photography when the analytical encountered when deploying them (especially in
output is intended to perform at a scale of 1:1,000 the case of fixed-wing SUA or kites) can outweigh
is pointless. Specifying such a GSD over a rough the benefits, and in some circumstances it can
grassed area would also be unhelpful, leading to be better to use a good compact or mirrorless
more data to be handled with no visible benefit to camera. Careful thought should therefore be
the product whatever the desired scale of output, given to the requirements of the end product
and the creation of unnecessarily large archives. before the acquisition and processing of images is
This holds true for all circumstances in which even started. However, it is always better to start
photogrammetry is used: the achievable densities with a slightly higher quality product in mind,
of point cloud and mesh can easily exceed the which can be reduced to the required output
requirements of the product and the capabilities density later, rather than a lower resolution
of the hardware employed. Although results product: if the data is not there to start with,
obtained using digital single lens reflex cameras it cannot be reliably interpolated.

Figure 2:
Overlapping images of a subject.

< < Contents 3

1 Basic Principles

This section provides an elementary introduction to some of the general principles

underpinning the photogrammetric process, and shows, in a highly condensed way,
how it works and why. Many of the operational principles derived from more traditional
stereo photogrammetry are equally applicable to more recent, highly automated
multi-image convergent photogrammetric developments that use a combination of
SfM and multi-view stereo (MVS) workflows. In practice, much of this will be hidden
from the user, but a better understanding of the basic processes involved will lead to
improved configuration of input imagery, more efficient processing and more
accurate results.

1.1 Basic procedure Image matching

Tie points on the images are matched (Figure 3)
The basic procedure for surveying most subjects, and the camera orientations deduced (Figure 4).
whatever the scale, is as follows.
Dense point cloud generation
Image capture A dense point cloud is generated, comprising
A series of overlapping images of the subject is all the possible matches (or a subset of all
taken (Figure 2). The example in Figure 2 is a piece the possible matches, depending on settings)
of carved stonework. between the images projected in 3D space
(Figure 5).

Figure 3
Tie points detected on the images.

< < Contents 4

Figure 4 (top) Figure 5 (bottom)
Interior and exterior orientations of cameras calculated. The dense point cloud, comprising all possible matches.

< < Contents 5

Figure 6 (top) Figure 7 (bottom)
A range of possible outputs. From left to right: DEM, Analytical products derived from the outputs shown
ortho-image and textured model. in Figure 6. Left: Slope analysis, Right: Multi-directional
hillshade.

Secondary product generation 1.2 The chief ray and the principles
A range of possibilities exists for the outputs, of intersection
including ortho-images, digital elevation models
(DEMs) and textured meshes. Figure 6 shows When an image is captured by a normal
examples from aerial imagery. Not all products camera, any point on that image represents
are suitable for all subjects. the convergence of many light rays. For
photogrammetric purposes, the ray of interest
Analysis and/or presentation of results is that which theoretically passes in a straight
Further analysis of the products can be carried line from the object point (A, B, C), through the
using other software (Figure 7), or the models can perspective (or projection) centre at the back of
be exported directly for visualisation purposes. the lens assembly (P), and onto the image plane
(I) at positions A’, B’ and C’ (Figure 8). This ray
The principles underlying each of these stages is often referred to as the chief ray (Fryer et al
are outlined in more detail in section 1.2 on 2007). If the interior and exterior orientations (see
The chief ray and the principles of intersection section 1.4 Interior and exterior orientation) of
and section 1.6 Measurement from single, paired multiple cameras are known, then the intersection
or multiple images. of the chief rays representing the same object
point in the images will define the position of the
object point in space (Figure 9). This process can

< < Contents 6

be applied across all, or a sample of, pixels for
which there are correspondents in other images.
It is important to note, therefore, that any part
of the subject that is not shown in at least two
images cannot be reconstructed.

1.3 Collinearity

Given the assumption that the object point,

camera perspective centre and image point are in
a straight line, in order to translate between the
two-dimensional (2D) image coordinate system
and the 3D ‘real-world’ coordinate system of the
subject, transformations must be applied. These
transformations are known as the collinearity
equations, and are based on an ideal camera as
shown in Figure 8, in other words one in which
there are no distortions and planarity (flatness)
of the sensor is assumed. In such a camera, there Figure 8 (top)
would be no geometric distortion from the lens Principle of the chief ray for any object point.
imaging system, and the transformation from 3D
object space to 2D image space is done using Figure 9 (bottom)
a perfect, central projection system. In normal Intersection of rays from two images defining object
points in 3D space.
use, however, there are several factors that
complicate the situation and must be accounted
for. In real cameras there are always geometric
distortions, which means that the image points
are slightly out of position according to the lens distortion parameters (k1, k2, k3, k4) and
idealised central projection. These deviations two decentring lens distortion parameters (p1,
from the ideal must be quantified, described p2), as well as determination of the principal
mathematically and compensated for. As well as distance (equivalent to the calibrated focal length,
lens distortion, other factors that can affect the which is the distance between the image plane
outcome include refraction and non-planar image and the perspective centre) and principal point
sensors. Refraction is not normally an issue in offset. As stated in Collinearity, the 2D image
most archaeological contexts, but comes into play coordinate system has its origin at the centre of
when oblique aerial images are shot at relatively the image. The principal point is the orthogonal
high altitude. In most digital cameras planarity of projection of the projection centre on the sensor,
the sensor can be safely assumed. and is not necessarily the same as the centre of
the image, hence the necessary computation
of principal point offset values. In a traditional
1.4 Interior and exterior orientation photogrammetric approach these values would be
derived through the process of camera calibration,
The model describing the geometric properties using a test field with objects of known position to
of the camera and lens system is known as the model the lens distortion. Calibration is discussed
inner or interior orientation, sometimes also in more detail below (section 2.1.1). In order to
referred to as camera intrinsics (Luhmann et compute these values correctly, it is essential that
al 2006). This includes modelling of the lens the original image is not cropped.
distortion, usually characterised by four radial

< < Contents 7

The interior orientation describes the parameters 1.5 The SfM–MVS
required to allow the principles of collinearity photogrammetric process
to be applied to distorted images. The exterior
orientation (also referred to as camera extrinsics) Now that some of the underlying principles have
describes the position (for example x, y, z) and been laid out, a slightly more detailed summary
attitude (roll, pitch and yaw, or omega, phi and of the steps in the SfM–MVS photogrammetric
kappa) of the camera’s projection centre when process is given.
the image was taken.
The SfM part of the process generates a sparse
The SfM approach, used, for example, by point cloud comprising tie points identified and
Bundler, Agisoft Photoscan, Photomodeler and matched across the input images. Other products
Pix4D mapper software, performs an automatic include the interior and exterior orientations for
calibration using, in the case of digital images, each camera, but these are rarely revisited by the
some of the exchangeable image file format (EXIF) end-user in most workflows. In order to construct
metadata in the image file as a starting point. the sparse point cloud, several steps are involved.
This defines the camera’s interior orientation
and simultaneously calculates the exterior Once images have been acquired and imported,
orientations using tie points identified on the the first step is the identification of features, or
input images in a process known as bundle interest points (IPs), on the images (see Figure 3
adjustment (a reference to the bundles of light for an example). The main requirement is that the
rays converging on the optical centre of each definition of IPs should have good repeatability:
camera). This process seeks to minimise the the same IPs should be detectable across images
re-projection errors between observed and under different lighting conditions and with
predicted image points. Unless a pre-calibrated different levels of image noise (Hartley and
camera model is used, it does this for every Zisserman 2003), a quality known as invariance.
image (or camera model) where the settings have There are several algorithms that can achieve this,
changed, allowing increased freedom in image for example the scale invariant feature transform
capture beyond traditional photogrammetric (SIFT; Lowe 2004), which is well known among
processes, such as the use of zoom lenses and photogrammetric developers. In addition to the
uncalibrated camera setups. In these situations, IPs, a similarly robust descriptor for each IP is also
camera calibration is undertaken by the software required, which describes a small area of pixels
‘on the fly’. around each IP, to facilitate matching.
Many IP detection algorithms generate this
The resulting ‘model’ will be in an arbitrary descriptor at the same time as the identification
coordinate frame and at an arbitrary scale if no of the IPs themselves. The number of IPs
formal control is available. If accurate control identified on each image is often set by the
measurements are available (measured points user; the default value in Agisoft Photoscan, for
rather than those generated automatically), they example, is 40,000 per image.
can be used to orientate and scale the output,
as well as provide refinements to the computed Next, the IPs are matched across the different
positions of the cameras and a check on the images. False matches are filtered out using an
overall accuracy of the model. outlier detection algorithm such as RANdom
SAmple Consensus (RANSAC; Fischer and Bolles
1981). Some software allows sub-selection of
only the best matches for each image. Once a
robust set of IPs has been identified and matched
across image pairs, the SfM algorithm needs to
estimate the interior and exterior orientations
for each image by combining all the relative
orientations of the image pairs in the form of

< < Contents 8

their fundamental matrices (Verhoeven et al during reconstruction than to follow the SfM–MVS
2013). Once complete, a technique called image workflow through to completion, produce the
triangulation is used to calculate the relative model independently and attempt to define a
position and orientation for each image in every coordinate system afterwards, as no refinement
pair. The overlapping pairs are then combined to the reconstruction parameters is possible at
to form a single block, the optimisation of which that stage.
is achieved by a bundle adjustment (see section
1.4 Interior and exterior orientation), so called Once the SfM part of the process is complete,
because it necessitates adjusting the bundles of the dense MVS reconstruction can be undertaken.
rays between each camera’s projection centre Now that the optical characteristics of the
and the set of projected 3D points until there is cameras and relative positions of the images
minimal discrepancy between the positions of are established, all possible IPs in each image,
the observed and re-projected points (the image including those with poorer repeatability than
distance between the initial estimated position the IPs used at the SfM stage, are calculated to
of a point and its ‘true’ or measured value) form a dense point cloud, which is similar in
(Verhoeven et al 2013). appearance to that generated by a terrestrial
3D laser scanner. There are many algorithms
In Agisoft Photoscan, for example, the IPs are available to do this, and different software will use
termed key points. Tie points (the sparse cloud different implementations. The dense point cloud
points seen in the model view after alignment) can then be used as the basis of a triangulated
are IPs (key points) that have at least two irregular network (TIN) or mesh, onto which
projections each: they are key points that have textures generated from the input images can be
been matched on two or more images and projected. The TIN can also be used to generate a
therefore have become potential tie points. When raster grid output suitable for use in a GIS.
a tie point limit is used, the software will use only
the most reliable tie points on each image to fit
the threshold set by the user (for example the 1.6 Measurement from single, paired
top 1,000 per image), which will result in a lower or multiple images
number of sparse cloud points chosen from only
the most reliable matches. Using very high IP and This guidance is principally concerned with the
tie point limits is rarely productive: it will result use of multiple overlapping images as opposed
in longer processing times and can also affect the to single images or stereo pairs. However, it is
accuracy of the alignment because less reliable important to recognise that using many images
IPs might be used in the matching process, will not necessarily increase survey reliability
resulting in less accurate tie points being selected. and can be surplus to the requirements of the
product, so it is often useful to consider the
The result of all this is a scale- and orientation- options of using fewer images as outlined in
free initial reconstruction. If a minimum of three the sections on Single image (1.6.1 below) and
control points (see section 2.3 Control) are Stereo pair (section 1.6.2). Further guidance on
introduced and used as constraints in the bundle these techniques can be found in other Historic
adjustment, they can be used to reduce further England/English Heritage documents (for example
errors in the reconstruction (such as the ‘dishing’ Measured and Drawn; see section 2 General
or ‘bowl’ effect sometimes seen as a result of considerations).
processing strips of aerial imagery; (see section
1.8 General workflow) and will also define a 1.6.1 Single image
coordinate reference system for the model. In One of the simplest ways to get measured
some software this is not possible, but Agisoft information from a single image is by using a
Photoscan and Pix4D mapper, for example, do process known as image rectification. As this
permit it. For accurate reconstructions, it is guidance is dealing with photogrammetric
better to integrate the control measurements processing, only an outline description of image

< < Contents 9

 so is most suitable for planar (flat) surfaces. If
the subject is three-dimensionally complex,
with considerable projections and recessions,
or undulating, you should consider using
an alternative technique. In single image
rectification, any image, even if free of tilt, will
exhibit displacements because of topographical
relief on the subject. Thus, any feature that is
either ‘below’ or ‘above’ the reference plane will
be misplaced and at the wrong scale because
of the central perspective of the image and
relief displacements. The greater the tilt of
the photograph relative to the main plane of
the surface, the greater the error as a result of
this relief displacement (also known as height
displacement/distortion when applied to
aerial images).

There are several limitations to rectified

photography, but in some circumstances it can
be an extremely effective and cheap solution.
Terrestrially, it is usually most applicable when
recording architectural facades, floors or ceilings,
and software designed specifically to deal with
these circumstances includes FARO’s Photoplan.
There are also a number of aerial applications
for rectified photography. In this case, if you have
access to photogrammetric software, generating
an ortho-image from more than one image (if
available) is preferable, as it takes account of
relief displacement (see section 1.6.2 Stereo pair),
although to some extent aerial image rectification
software such as Irwin Scollar’s Airphoto and
Figure 10 John Haigh’s Aerial use existing height data from
Stages of single image rectification. other sources to mitigate this. The principles of
single-image rectification are shown in Figure
10. In this case, software running inside a CAD
package is used to match points in the image to
rectification will be given here. Rectification surveyed points, a rectified image produced, and
involves the projective transformation of a detail traced from it. The smaller the amount of
single tilted image to a plane to remove the tilt distortion present in the original image, the better
displacements, optionally including an estimation the rectification is likely to be.
of lens distortion parameters. Thus, an image
with perspective distortion, as taken by a normal Images can be rectified to some extent at the
camera without special lenses, is re-projected to point of capture by using rising front/perspective
remove that distortion as far as possible, and in a correction/tilt-shift lenses, but these present
plane that is parallel to the object plane. real problems for photogrammetric processing.
These lenses are typically used in architectural
As rectified photography uses only one image, photography to remove or reduce the effect of
it is effectively a 2D recording technique and perspective, and in this application are usually

< < Contents 10

of relatively short focal lengths (up to 35mm; 1.6.2 Stereo pair
Figure 11). The main problem is that the position When two overlapping images are available
of the perspective centre is moved relative to (a stereo pair), a digital surface model (DSM)
the principal point (by a physical offsetting of can be derived in the overlapping area using the
the lens assembly), and this is extremely difficult principles of intersection described in section 1.2
to compensate for in the photogrammetric The chief ray and the principles of intersection.
processing of such imagery, given that the offset A typical image arrangement for a stereo pair is
is typically not recorded in metadata (certainly not shown in Figure 12. Reliable measurements can be
in pre-digital imagery taken with such a lens, or taken in the overlapping area between the
with a lens that cannot supply such data fitted to two images.
a digital camera). When using historical imagery
it can sometimes be difficult to spot whether A single image rectification aims to remove the
such a lens has been used, and errors in tilt from the input image but is reliable only for
processing can be significant and the accuracy the parts of the subject that coincide with the
of the product severely compromised. The use of rectification plane. The DSM derived from two
such imagery for measurement is therefore not overlapping images allows the effects of relief
recommended; clues in images that these types displacement and other geometric distortions
of lenses have been used include tall buildings to be taken into account, so that the resulting
that have parallel vertical lines rather than the ortho-image is free from such distortions and can
perspective distortion typical of using a ‘normal’ be used like a map (with aerial images) or a plan/
lens, as seen in Figure 11. elevation (with terrestrial images), irrespective

Figure 11
Images taken using a ‘normal’ lens (left) and a tilt/shift
lens (right) to remove perspective distortion.

< < Contents 11

of the 3D variation of the subject matter. Images 1.6.3 Multiple images
forming a stereo pair are usually taken parallel Just as a stereo pair can be used to compute the
to one another. In traditional photogrammetry, 3D positions of matched IPs in the overlapping
each pair forms a stereo model, although the right area, in current photogrammetric software many
image of a pair in one model would usually also images can be used. In most cases a single stereo
be the left image in the next pair if the imagery pair will not provide the coverage necessary to
was taken in a strip configuration, for example. allow reconstruction of the entire subject. Under
A comparison between a rectified image and an these circumstances more images are required,
ortho-image is shown in Figure 13. and nowadays photogrammetry is no longer
restricted to the former stereo pair constraint of
In Figure 13, it can be seen that although parallelism of the input images. Oblique (both
measurements can be taken from the central area horizontally and vertically) and convergent
of the rectified image (between the buttresses, images can be used as well as parallel images,
above the plinth and below the window), which and this provides a number of additional benefits.
has been chosen as the rectification plane, the A convergent image configuration minimises the
systematic errors caused by inaccurate estimation
of lens distortion characteristics (Chandler 2010;
Figure 12 Wackrow et al 2008) and can also provide 100
Typical camera configuration for a stereo pair. per cent overlap of image pairs if required, thus
permitting more efficient subject coverage and
allowing useful images to be taken in situations
where ‘normal’ stereo photography would be
rest of the image is still subject to perspective difficult to apply. Regarding aerial images,
distortion (for example where the buttresses current photogrammetric software permits the
project from the wall face), and these parts of the use of oblique images from flights that circle the
image are therefore displaced and cannot be used subject, rather than flying in the more typical
for measurement. In the ortho-image, in contrast, overlapping swaths used for aerial mapping; the
the effects of perspective have been removed and addition of oblique images to a typical vertical
measurements can be taken reliably from any set can significantly increase the accuracy of the
part of the image. results (Nocerino et al 2013; Wackrow et al 2008).

Figure 13
Comparison of a rectified image (left) and an
ortho-image (right) of the same subject. The rectified
area is shown with dashed line.

< < Contents 12

When using digital cameras, extra images can
be taken with ease to increase data redundancy
while simultaneously strengthening the geometric
configuration. Figure 14 shows the difference
between stereo pairs and a highly convergent set
of images.

When dealing with multiple images, the bundle

adjustment process, as described in section
1.4 Interior and exterior orientation, is used to
optimise the 3D reconstruction and refine the
interior and exterior orientations such that re-
projection errors across the whole model are
minimised. When the 3D positions of the tie points
are estimated, those points are re-projected onto
the images: the difference between the detected
and re-projected point position on an image is the
re-projection error, and is dependent on both the
quality of the camera calibration estimates and (in
the case of manually marked points) the accuracy
of the marking. It thus provides a good indication
of the accuracy or otherwise of the reconstruction.

1.7 Sources of error

Figure 14
There are two main sources of error in Convergent (top) and ‘normal’ (bottom) stereo case.
photogrammetric processing that can be
compensated for. You need to be aware of both.

1.7.1 Systematic errors

These can be caused in a variety of ways, but Problems with sensors are more difficult
are mainly concerned with factors that affect the to quantify and correct but can, in some
interior orientation. They include the following. circumstances, be helped by calibration (Fryer
et al 2007). In most digital cameras, planarity
1.7.1.1 Sensors of the sensor can be safely assumed; incorrect
„ Non-planar sensors interior orientation parameters can be resolved
during bundle adjustment, and these can also be
„ Physical errors in the pixel geometry of improved to some extent by calibration.
the sensor
1.7.2 Mistakes
„ Non-perpendicularity between the plane of Mistakes usually involve either incorrect matching
the sensor and the lens axis of points during automated alignment or the
incorrect identification and/or measurement of
1.7.1.2 Other control points. In the former case adjustment
„ Incorrect lens distortion estimates can be repeated after manually orientating the
problem images, while in the latter case the
„ Incorrect positioning of the principal point misidentification can usually be found quickly
and rectified.
„ Refraction

< < Contents 13

1.8 General workflow refinements to the process, many of which will
increase the likelihood of generating accurate
The general workflow for SfM photogrammetric outputs with verifiably good metric performance.
processing is outlined here. At all stages metadata The order in which some parts of the workflow
is important; some of this will be generated have to be undertaken can vary depending on
automatically by the software employed the software being used, as do the refinement
(including processing and accuracy reports, options available.
if available), while other elements should be
generated by those working on the project. All images should be checked before passing them
through the workflow, primarily to remove those
„ Planning that are of poor quality, usually those that are
comprehensively out of focus or exhibit significant
„ Reconnaissance motion blur as a result of either incorrect camera
settings or the use of frames grabbed from video.
„ Image acquisition Some software has image quality assessment
functionality; none of their methods are perfect,
„ Image pre-processing however, and a visual inspection of the inputs
is always advised before processing starts.
„ Image import Individual images that cannot be correctly
aligned can be removed, provided there is
„ SfM, calculation of interior and exterior sufficient redundancy in the inputs, or can be
orientation, identification of IPs across the manually aligned with the introduction of local
image set, formation of sparse point cloud image control points. Additionally, problematic
based on those IPs areas of images can, in some software, be masked
to remove features that do not need to be
„ Incorporation of control data, reconstructed, for example sky, logos, fiducial
alignment optimisation marks or moving elements in a scene. This can
save a considerable amount of time later in
„ MVS, formation of dense point cloud by the process.
parsing all images and projecting most
of the pixel data contained in them as 3D Sparse point clouds can, in various software, be
points, provided they can be matched and filtered to remove points with high re-projection
identified in at least two of the input images errors or reconstruction uncertainty. In general,
the Root Mean Square Error (RMSE) of the
„ Dense point cloud editing (optional) re-projection errors should be below 1 pixel; the
smaller the value, the better the accuracy of the
„ Generation of other outputs (high-resolution reconstruction. Although filtering for re-projection
mesh, ortho-images, DEMs, etc) errors is an iterative process, you should not
carry out the filtering step more than three times,
„ Further processing and analysis of those as this can introduce a ‘stepping’ effect into the
outputs in other software (CAD, GIS, etc) model (Figure 16). You should also avoid removing
too many points during filtering, as reconstruction
„ Presentation can then become compromised or impossible.
If the input images are producing very large
„ Archive re-projection errors that cannot be mitigated,
it is likely that they are either of poor quality or
Figure 15 shows the basic elements of a close- the camera calibration parameters cannot be
range photogrammetric workflow in graphical accurately determined.
form. At all stages there are several possible

< < Contents 14

Process Outputs Metadata

Project planning and reconnaissance Project documentation Project Metadata

No calibration Calibrate Cameras Calibration Report(s) Camera Metadata

Coordinate Frame
No control Establish Control (from measured points Control Metadata
and /or object constraints)

Acquire Images Digital images Image Metadata

Image Pre-processing Processed digital images XMP Metadata

Processing

SfM/MVS model generation

Metadata

Partially/
Uncontrolled
deliverables
Fully controlled
deliverables
Object/surface model

Texture

Dense Point Cloud

{
Deliverables

Metadata

2D raster DEM

Ortho-image

Images, Deliverables and

Metadata to Archive

Figure 15
Basic elements of the photogrammetric workflow.

In cases where coordinated control point data is This mitigates the effects of inaccurate camera
available, the whole process can be refined, for calibration estimates and also greatly increases
example in Agisoft Photoscan Pro, by building the accuracy of the reconstruction, depending
a low-accuracy dense point cloud and a low- in part, of course, on the accuracy of the control
resolution mesh (in order to enable semi- measurements. The newly refined alignment
automated placement of control); after carefully can then be used as a basis for re-initiating the
positioning all control points and assigning next stages of the workflow, as the previously
their coordinates, optimisation of the image generated dense cloud and mesh will have been
alignment can be carried out using those values. rendered redundant and removed.

< < Contents 15

 In Figure 17, the ‘bowl’ effect is caused by
inaccurate initial camera calibration estimates at
the image alignment stage. This can be mediated
by either loading the correct camera calibration
values before alignment (if a calibrated camera
system is used) or by using ground control points
(GCPs); the software will then take into account
the additional constraints introduced by the
GCPs when the alignment is optimised. The use
of GCPs to increase reconstruction accuracy is
recommended even if a calibrated camera is used.

Figure 16
Over-use of filtering, leading to a pronounced
‘stepping’ effect in the mesh.

Figure 17
‘Bowl’ effect caused by inaccurate initial camera
calibration estimates. The problem is not immediately
visible in plan (top), but the side view (bottom) shows
the effect.

< < Contents 16

 Software formal control in the processing of the data,
and furthermore do not allow optimisation
Software available for structure from motion of image alignment based on control data.
(SfM)/multi-view stereo (MVS) processing Instead, the SfM sparse point cloud is built,
can be grouped in many ways: here it is the dense cloud then constructed, and this
divided by cost into free and commercial product (or a mesh derived from it) positioned
packages. A useful comparison of the metric using an affine transformation incorporating
performance of a variety of these packages the control points surveyed in the field. This
using the same data is given in Remondino can be achieved in other open-source software,
et al (2012). It should be noted that this is a such as Cloud Compare or Meshlab. In many
rapidly developing and expanding market: cases this does not present any difficulties,
any references given here are likely to but serious problems can be encountered,
be superseded extremely rapidly, as new especially when working over large areas, if
companies move into the area from a wide working with projected coordinate systems
variety of commercial perspectives and [that is those gathered using global navigation
applications. It should also be noted that this satellite systems (GNSS) systems, or a total
section is in no way an endorsement of any station theodolite (TST) when using a scale
of the packages described, but a summary of factor of anything other than 1], as a degree
some of the more popular software currently of distortion is introduced into the control
(2017) being used in the archaeological sector measurements that cannot be compensated
in the UK. for unless the software is capable of dealing
with projected data. The net result of this is
Free solutions are generally either released that the control measurements cannot be used
under the GNU public license, such as Bundler, properly, and the results will not align well
PMVS2 and the GUI ‘wrapper’ for them, Visual with data derived from other sources or when
SfM (Wu 2011), or offer cloud-based solutions overlain on, for example, an Ordnance Survey
using a system where the user uploads map in a geographic information systems (GIS)
photographs that are processed and the or computer-aided design (CAD) system.
model returned, such as KU Leuven’s Arc3D
and many others. These are discussed in Reliance on the SfM process alone can also
more detail below. generate errors in some data sets (for example
the bowl effect shown in Figure 17). These
Open-source software can be compensated for to some degree by
There are many advantages to using using calibrated cameras, but the best way to
open-source software, among them cost, fix them is to optimise the alignment based
development options through access to the on either accurate camera or ground control
source code and ownership of data and format point (GCP) coordinates. Without the ability
(Green et al 2014). However, using this system, to incorporate control measurements into
getting from the starting material (the images) the SfM/MVS workflow at the time of model
to products [digital elevation models (DEMs), creation, these types of errors cannot be
ortho-images, etc] requires the use of several mitigated by most open-source solutions at the
different pieces of software at different stages time of writing, although the functionality is
(Green et al 2014). found in many commercial packages.

The freely available desktop solutions, of One potential advantage of the Bundler/PMVS2
which perhaps the most user-friendly is workflow is that it involves processing all of the
Visual SfM, do not generally allow the use of data on a local machine. This means that the

< < Contents 17

 process can be controlled by the end-user, and less expensive end of the market. Some
elements of the workflow changed according to solutions focus almost entirely on the aerial
the requirements of the job at hand. A potential survey sector as their primary goal (such as
disadvantage, however, of all systems (free or SimActive’s Correlator 3D; although these
commercial) that manage the data locally is are not cheap, the potential return on the
that large data sets can take a very long time to investment is good if they are being used in a
process, depending on the hardware available. commercial environment on a regular basis.
The SfM/MVS workflow is computationally More versatile solutions include the ability
intensive, especially with large numbers of high to work with terrestrial imagery as well as
resolution inputs. It is often difficult to process aerial imagery, deal with highly convergent
effectively on a laptop, for example, where image sets, and produce full 3D models from
resources are typically insufficient to obtain a variety of inputs. The most popular include
good results in a reasonable timeframe unless Agisoft Photoscan, Pix4D mapper, 3DF Zephyr
few, or relatively low resolution, images are and RealityCapture; the latter additionally
used. One potential solution to this problem is allows the direct integration of data derived
cloud processing, using many fast computers from both photogrammetry and 3D laser
elsewhere to perform the ‘heavy lifting’ parts of scanning. Different licensing models are
the operation, and to evaluate and analyse the sometimes available (for instance, licenses for
end results locally. Pix4D mapper may be rented for a monthly or
annual period in lieu of an outright purchase,
Free cloud-processing solutions and RealityCapture currently offers a rolling
Such solutions include Autodesk’s Recap license model that is considerably cheaper
Image or Recap 360, although both also offer than the full purchase price) that can make
versions with more functionality at a cost, and these solutions more affordable for those
KU Leuven’s Arc3D, that are all useful for cloud- on limited budgets. Like the open-source
based 3D reconstruction. They do not down- solutions, they allow local processing of data,
sample images, and the models produced, and therefore increase the amount of control
although still arbitrarily scaled and orientated, that the surveyor has over the data and the
have relatively good metric performance and processing workflow as well as integrating
are beautifully textured. In many situations the external control data. Some, like Pix4D mapper,
product compares very favourably with that additionally offer a cloud-processing option for
generated using commercial packages. projects, while others, like Agisoft Photoscan
Pro, offer the capability for network processing
It is also important to note that copyright of of projects, both solutions going some way
some of the results from cloud-based free to obviating the need for very powerful
software may reside with the software provider computing resources in-house.
and is in many cases not licenced for use in a
commercial context. Commercial cloud-based solutions, such as
Drone Deploy, offer end-to end processing
Commercial packages solutions that can optionally include flight
There are many commercial photogrammetric planning, and which offer processed outputs
packages available, at a range of prices. including 3D models, DEMs, ortho-images
Outside academic institutions, however, and normalised difference vegetation index
relatively few of the more expensive solutions (NDVI) composite images (if appropriate
(for example BAE Systems’ Socet Set or imagery is uploaded).
Hexagon's Erdas Imagine) are used in UK
archaeology, so the focus here is on the

< < Contents 18

2 General
Considerations
2.1 Capturing the initial data retains versatility across many photogrammetric
scales of operation, in part through the
In photogrammetry, the quality of the output application of different lenses. The use of metric
is almost wholly dependent on the quality of the cameras in UK archaeology is not widespread, so
input. Poor photography will inevitably lead to their use will not be considered in any detail here.
inaccurate results, so time spent familiarising
yourself with the camera you intend to use, and Broadly speaking, DSLRs will provide better
considering the best image configurations for results than compact cameras, and these will in
the subject, is seldom wasted. This section will turn provide better results than ultra-compact
summarise some of the main issues; there are cameras (Wackrow 2008) or mobile phones, for
several excellent resources for more detailed many reasons as discussed in Capturing the
expositions of how cameras work and how this initial data. In some cases, the camera that can
is relevant to the photogrammetric process (for be deployed may not be the best from an image
example Stylianidis and Remondino 2016, quality point of view, but may be necessary
127–251). because of weight or bulk restrictions, a situation
commonly found when using fixed-wing SUAs,
In general, optimal exposure for photogrammetry, kites or masts. In any case, the best quality
as in ‘normal’ photography, involves a balancing camera available should be chosen for the job,
act between aperture, shutter speed and sensor which is not necessarily the camera offering the
sensitivity. The aim is to produce clean, sharp highest pixel count. Unless money is no object,
images of the subject. You should aim to use you should not buy the newest, or most high-
the fastest shutter speed that conditions allow end, camera on the market, as these are normally
(to reduce the chance of blurring), the lowest very expensive and prices usually come down to
ISO setting possible (to reduce image noise) and affordable levels within a year or two. Money is
the optimum aperture to retain sharpness and much better spent on good-quality lenses, which
appropriate depth of field (often between f/8 often retain usability with newer camera bodies
and f/11) across the subject. over many ‘generations’.

There is no ‘best’ camera for all photogrammetric Some software can process ‘spherical’ imagery
work, although a single good-quality camera taken with, for example, a Spheron VR camera,
can be a far more versatile sensor in a variety iStar or other 360 degree imaging systems. While
of situations than, for example, a much more these can provide excellent coverage and overlap
expensive 3D laser scanner. Different types of in an indoor setting, when used outdoors a large
laser scanner perform well at some ranges and number of the pixels may be imaging sky and
tasks but not at others (3D Laser scanning for ground, reducing considerably the number of
heritage), but a single good-quality camera effective pixels available for modelling the subject

< < Contents 19

(Figure 18). Unless using in-camera high dynamic the past medium- and large-format metric film
range (HDR) imaging, interior illumination will cameras often used a partial vacuum to suck the
often be challenging and can limit the usability of film flat. In modern digital cameras, it is generally
lower cost 360 degree cameras. assumed that the sensor is planar. Thus the only
factors left that can affect output are refraction
2.1.1 Calibration (as mentioned in section 1.6.1 Single image),
Although much of the photogrammetric software not usually an issue in archaeological work) and
currently in use in archaeology does not require lens distortion, and it is these lens distortions that
a pre-calibrated camera (with calibrations the calibration process seeks (in part) to model
being calculated for each camera during bundle and mitigate.
adjustment), in most cases more accurate
results can be obtained using a setup in which A variety of methods are available for camera
the distortion parameters are consistent and calibration, ranging from professional 3D
measured in advance with a calibrated fixed test fields (which are beyond the means of
focal length (prime) lens. Adequate results can most but can be commissioned) to relatively
be gained from very variable input photography straightforward solutions that usually involve
using, for example, a zoom lens on a camera photographing a 2D test image from a variety of
set up to adjust all settings fully automatically angles in order to provide an approximate lens
for ‘optimal’ exposure. Much of the time, the model. In all cases, the aim is to estimate (at a
software will be able to estimate successfully the minimum) the radial lens distortion parameters
interior orientation values for the camera for each (k1–k4) and decentring lens distortion parameters,
exposure with different settings, and to apply also known as tangential distortion (p1, p2), thus
these during the reconstruction phase. enabling a reasonably accurate estimation of
the principal distance (calibrated focal length)
The collinearity equations assume that the image and the principal point, together providing the
point, projection centre and object point are in interior orientation. Once all of these parameters
a straight line, and that the image is formed on a have been determined, they can be applied
plane. This last assumption is the reason why in to the images in such a way that the idealised

Figure 18
Outdoor use of a spherical camera.

< < Contents 20

assumptions of the collinearity equation can be

used to reconstruct 3D points from 2D inputs
when combined with an estimate of the exterior
orientation (position and attitude) of the camera.
In some very cheap cameras and mobile phones
it cannot be assumed that the sensor is in fact
perpendicular to the lens axis because it may be
glued in place (Bradski and Kaehler 2008), leading
to severe distortion. In this case the calibration
routine should also take into account skew and
different values for Fx and Fy (the focal length in
x- and y-dimensions measured in pixels); in Agisoft
Photoscan, for example, these coefficients are Figure 19
estimated during camera alignment optimisation Prime lens taped to focus at 1.2m.
(A Pasumansky, pers comm).

Before starting a calibration, settings appropriate

for the project should be chosen. The aperture Examples of test fields used by different software
of the lens should be stopped down to give packages are shown in Figure 20. Calibration
sufficient depth of field: choose the sharpest should be re-done if:
aperture setting for your lens (usually around f/8),
which will vary depending on the requirements „ settings have been deliberately changed
of the job, and calibrate using this value. Using (for example a new focal length selected)
calibrated fixed focal length (prime) lenses will
involve setting the focusing distance appropriately „ settings have been accidentally changed
(typically you might choose to focus at infinity, (for example the camera has been dropped,
or perhaps for closer range work at 1m or 2m) knocked or treated roughly)
and then preventing this from changing (using
electrical tape or a locking screw, if available). „ the lens has been removed or replaced
In Figure 19, the focal length has been set to
infinity, and a note of the settings made on the If used, calibration should be undertaken
tape. The camera should be set to manual mode, approximately every 12 months, or before starting
and electronic assists such as auto-focus (AF) a major project. With the correct setup, it can be a
and image stabilisation should be switched off, relatively rapid process.
in many modern systems on both the camera
and the lens. At this point, the instructions for 2.1.2 Resolution and sensor size
whatever calibration software you are using The resolution and sensor size of different
should be followed. cameras varies widely, and has a significant
impact on the quality of the images produced.
Some photogrammetric software is very good Medium- and large-format sensors are not very
at estimating robust interior orientations, and widely used in UK archaeology, and their use will
may be better at it than some of the cheaper (or not be considered here.
free) calibration software available, although
the latter can provide useful starting points for Generally, the larger the sensor and, to some
parameter estimations. There is no standard extent, the larger the individual pixels on it,
format for camera calibration results, so it can be the better the image quality will be. You will be
problematic moving the results from one piece of able to distinguish and model more detail in a
software into another; photogrammetric software high-resolution image than in a low-resolution
packages typically only allow the import of some image because the same parts of the subject
results from other packages. are represented by more pixels in the higher

< < Contents 21

resolution image, assuming the image scales them. From this, it can be seen that a compact
are the same. camera can have a sensor size only 20 per cent
that of a 35mm full-frame sensor. If the pixel
Very small sensor arrays with densely packed count matches that of the larger sensor, then
pixels will generally produce noisier images, and the individual photo diodes on the sensor of the
be more likely to exhibit undesirable artefacts compact camera will be much smaller than those
and optical effects. DSLRs will usually have larger in the full-frame camera in order to fit on the
sensors than compact cameras, which in turn smaller sensor. Smaller pixels sacrifice a larger
have larger sensors than ultra-compact cameras proportional area to secondary circuitry on the
or mobile phones. Larger sensors come at a sensor and their more limited light-gathering
price, however. Not only do they cost more, but capacity requires additional signal amplification,
the cameras housing them are larger, the lenses resulting in higher signal noise. Thus closer
are generally larger and they weigh considerably pixel spacing on sensors can lead to decreased
more, which means that, for example for use with sharpness, a smaller dynamic range (which can
a SUA, a larger platform is needed to carry them. lead to problems with clipped highlights), less
colour saturation, increased chromatic aberration
Table 1 gives some typical values for a variety and consequently lower overall image quality. As
of sensor sizes and the cameras that can use an example, a Nikon D3X with a 24MP full-frame

Sensor Dimensions Area (mm 2) Percentage of 35mm Typical cameras and approx
name (mm) (approx.) full-frame (approx.) resolutions (mega-pixels, MP)
(approx.)

35mm 36×24 864 100 Nikon D3X (24MP), D800 (36MP)

full-frame Canon EOS 5D mark III (22.3MP)
Leica M (24MP)
Sony Alpha 7R II (42.4MP)

APS-H 28.7×19.1 548 63.45 Canon EOS 1D (inc. marks II–IV) (8.2MP
for mark II)

APS-C 23.6×15.7 370 43 Nikon D300 (12.3MP), D3000 (10.2MP),

(Nikon DX) D7100 (24.1MP), D80 (10.2MP), D70
(6.1MP)

APS-C 22.2×14.8 329 38 Canon EOS 7D (20.2MP), EOS 60D (18MP),

(Canon) EOS 50D (15.1MP)

4/3” 17.8×10 178 20.6 Panasonic Lumix DMC-L10 (10MP)

Olympus E5 (12.3MP)
Leica Digilux 3 (7.5MP)

1/1.7” 7.6×5.7 43 5 Canon Powershot G12 (10MP)

Ricoh GR (16.2MP)
Nikon Coolpix P7100 (10MP)
Panasonic DMC-LX5 (10MP)

1/2.3” 6.2x4.6 28 3.25 Sony Cybershot DSC HX50

Table 1
Typical values for sensor sizes.

< < Contents 22

sensor has a pixel area of 35.05µm 2 and a pixel
pitch (size) of 5.92µm, whereas a Sony Cybershot
DSC HX50 with a 20.4MP 1/2.3” sensor has a pixel
area of 1.39µm 2 and a pixel pitch of only 1.18µm.

As new cameras and sensors are released at a

rapid rate, the data in Table 1 will quickly become
out of date, but the principle being demonstrated
will remain relevant.

You can typically expect poorer results when

using cheap equipment and it is often a false
economy to do so. While it is acknowledged
that the imagery from very light ‘hobbyist’ SUA
configurations will process, the metric quality
of the results, as well as the resolution of fine
detail, is very often at the edges of what is
considered acceptable, as only cameras that are
light enough can be carried on such platforms.
Some cameras in this category, notably those
primarily designed for sports and recreational
use (being helmet or handlebar mounted), exhibit
high degrees of image distortion (Figure 21;
in this case made clearer by the horizon line),
considerable chromatic aberration in certain light
conditions, and were certainly not designed with
photogrammetric outputs in mind. That said,
some software manufacturers are building in the
capability to deal with imagery derived from such
cameras, given their popularity, robustness and
light weight.

2.1.3 Focus and sharpness

The sharper the image, the better it is for
photogrammetry, although there are caveats
about processing images to increase sharpness
(see section 2.1.10 Image enhancement).
Sharpness should be a result of good-quality
lenses and optimal exposure rather than image
enhancement. Focus is equally important: out- Figure 20
of-focus shots will process very poorly, and can Examples of different camera calibration images.
lead to serious errors being introduced into a
data set. It is well worth spending time checking
your images before running them through
the photogrammetric process. AF lenses in an
uncalibrated setup can sometimes be problematic
in this regard, and you should check whether
the correct focus point has been selected,
and adjust it (by switching the lens to manual
focusing, for example) if necessary. If you are

< < Contents 23

Figure 21
Severe image distortion from a recreational camera.
© Skyline images

using a calibrated lens, the focus will remain fixed 2.1.4 Depth of field
and any AF functions must be turned off. This is Depth of field is controlled by the aperture
particularly important if you are shooting under settings. The aperture is a hole of variable size
circumstances that will be difficult to replicate, that controls how much light is let into the
or at a site where access is limited and a return camera during an exposure. A camera is only able
visit will be problematic. Some photogrammetric to focus its lens at a single point, but there will
processing software permits the masking of areas be an area that stretches in front of and behind
of images that are out of focus before processing this point that still appears sharp. This sharp
begins, and if possible this option should be used. area represents the depth of field and is not a
Although resolution has a bearing on the quality fixed distance, as it is controlled by the size of the
of information that can be gained, it should be aperture. It can be described as ‘shallow’, where
realised that poor-quality lenses will yield results only a small zone around the focus point appears
that lack sharpness whatever the resolution of sharp, or ‘deep’, where a larger proportion of the
the camera used. Depending on the subject being image appears sharp. Larger apertures (smaller
imaged, the area of the image that is in focus is f/ numbers) lead to a shallower depth of field,
heavily influenced by depth of field. whereas smaller apertures give a deeper depth
of field (Figure 22). This can sometimes present

< < Contents 24

problems when shooting in low light conditions f2
where a larger aperture is desirable, especially
when a tripod cannot be used to allow longer
exposures to compensate. In these situations,
increasing the sensor sensitivity (ISO values) is
often the only other option. Most lenses have an
optimum aperture setting for minimising lens
distortion, usually between f/8 and f/11, but the
depth of field will increase across the whole range f4
of apertures. If you are photographing largely flat
surfaces this does not present much of a problem:
use whatever works best under the circumstances,
and much wider apertures can be safely used,
with the additional benefit that ISO values can
be reduced, resulting in less noisy images. It is
sometimes the case, however, that you need to
increase the depth of field (to keep more of the f8
image sharp) by decreasing the aperture to, say,
f/16 or smaller, in which case you will almost
certainly need to use a tripod to compensate for
the correspondingly longer exposures necessary,
especially in poor light.

Depth of field becomes increasingly important

the more ‘3D’ the subject is, or the more oblique f11
the images of the subject are: anything that is
not in focus usually has to be masked from the
input photographs or they cannot be processed
accurately. This is less of an issue in high-level
vertical aerial photography, where the ground
surface is relatively planar compared with the
camera under most circumstances, but becomes
more of an issue with lower level SUA or kite f16
photography, especially when that is oblique,
and can have a serious impact on some terrestrial
projects (Figure 23). It can also present difficulties
when using lenses that have very narrow depth of
field characteristics (such as section 3.4.1 Macro
lenses). Ultimately, you need to ensure that you
have as much of the subject as possible in focus
when the exposures are made. f22

Figure 22
Different aperture settings showing the effect on
depth of field.

< < Contents 25

 2.1.5 Film speed and sensor sensitivity (ISO)
ISO values (the approximate equivalent of ASA
film speed when using an analogue camera)
reflect the relative sensitivity of the sensor to light
(Figure 24). In general, ISO values should be kept
as low as possible to reduce noise in the image.
Increased noise at high ISO values is a well-known
phenomenon, especially in older digital cameras,
and can lead to significant degradation of the
derived products.

In certain circumstances, for example when

shooting from a mobile aerial platform, hand­
held device or in high winds, faster shutter speeds
are essential to avoid motion blur and to retain
sharpness in the outputs. In these conditions,
apertures will often have to be opened up (lower
f/ numbers selected) to allow more light to reach
the sensor during the exposure, and ISO values
may need to be set higher to result in a properly
exposed image. In these situations, ‘fast’ lenses
(those that permit wider apertures to be used)
are helpful, as they are in low light conditions,
but wider apertures are used at the expense of
depth of field.

When a stable platform (for example a tripod)

is available, smaller apertures and lower ISO
values can be used to ensure adequate depth of
field across the subject and a minimum of image
noise. Most cameras have the functionality to
use either shutter priority or aperture priority.
Figure 23 In the former case, the user sets the desired
Depth of field in an applied case. shutter speed and the camera adjusts the ISO and
aperture values to gain an optimal exposure, and
in the latter the user sets the desired aperture
and the camera adjusts the ISO values and shutter
speeds accordingly. Adjusting aperture is not
an option if a calibrated lens is being used. If
you are using a tripod in low-light conditions it
is possible on many DSLRs to use a mirror lock­
up mode, which further reduces the chances of
camera shake during a longer exposure, especially
when combined with a delay timer or wireless
remote trigger. Mirrorless cameras can be used
to mitigate this problem, and in some cases offer
an electronic front and rear curtain to avoid any
movement during the exposure process.

< < Contents 26

 ISO50 ISO100 ISO250 ISO400 ISO800 ISO2000

Figure 24
Effects of different ISO settings. Shutter speed and
aperture are constant throughout.

The use of external lighting rigs is advised if shooting very clean images even at high ISO
possible, especially in low-light indoor conditions, values. Where previously, and still if using older
to allow clean exposures in otherwise trying cameras, the advice was to stick to ISO 400 or
conditions. If cameras rely on a fully automatic below, it is now possible, if circumstances dictate,
setting, they often choose exposure settings with many modern cameras, to use much higher
that could have been improved by manual values with relatively little image noise penalty.
intervention, as in the aerial example shown in
Figure 25. In this instance, the shoot was taken on 2.1.6 Lighting
an overcast and very windy day, with a compact In contrast to ‘normal’ photography,
camera on full automatic settings (necessitated relatively flat lighting is generally preferred
by the firmware on the SUA platform being used). for photogrammetric purposes. Typically,
To compensate for low light and the high shutter lighting is used to emphasise texture; for
speed necessary to avoid blurring (1/2000s), the aerial archaeological photography, images are
aperture was opened wide (f/2) and ISO values often taken primarily to emphasise features
increased to 1600, resulting in considerable noise on the ground, with the sun at a low angle; for
in the final image, which in turn led to low-quality architectural photography, images are taken to
results. When re-flown with a better camera on elucidate details on a building or structure. Areas
a more stable platform, there was significant hidden in deep shadow may not yield the best
improvement in the imagery (Figure 26). photogrammetric reconstruction results, and
overexposed areas can have a negative effect on
In general, ISO values should be kept as low as the outputs. If ortho-images or model textures are
is practically possible under the circumstances, part of the desired output, it is best to try to avoid
although some newer cameras are capable of significant changes in ambient lighting conditions

< < Contents 27

 during a shoot. Although the metric properties of
the output should not be affected too badly (on
the understanding that exposures are not heavily
compromised by changes, for example, with a
good deal of over- or underexposure), the quality
of blended textures and computed pixel values
for the output points may be adversely affected.
There are ways round this when the changes are
relatively small (for example Agisoft Photoscan
offers average, minimum and maximum options
for texture generation in addition to the normal
blending modes and a facility for colour
correction, although this can incur a significant
time penalty, especially when processing large
numbers of images).

Try to avoid using the in-built flash on a camera,

as it is highly directional and the lighting on the
subject therefore changes dramatically between
exposures. It is far better to use ambient light, or if
possible to light a scene or subject using external
lighting rigs (Figure 27), to maintain a constant
light over the scene during capture. A mast can
be seen to the right of the image in Figure 27:
this was used to get images perpendicular to the
Figure 25 (top) wall paintings in addition to the shots from the
High ISO values producing noise in an aerial image tripod. If using external lighting sources, attempt
taken with a relatively cheap camera from a
to light the scene as evenly as possible and try
fixed-wing SUA.
to avoid pockets of deep shadow or highlights
Figure 26 (above) on the subject. Some LED lighting rigs allow you
The same area re-imaged with a better full-frame to alter the colour temperature as well as the
mirrorless camera mounted on a multi-rotor setup.
intensity of the light during a shoot. If colour
reproduction is important (as it nearly always is
in photogrammetric projects), colour reference
cards should be used (Figure 28) and camera
white balance settings adjusted according to the
prevailing conditions, Most cameras offer the
option of setting a custom white balance and this
should be used when the accurate rendition of
colour is important.

2.1.7 Lenses
Better results will be obtained by using better
quality lenses. Even on a high-quality camera,
poor lenses will yield images that lack sharpness
and clarity. Cheap lenses are almost always a false
economy for accurate photogrammetry, although
Figure 27 it is acknowledged that results can be obtained
Lighting rig in use while photographing wall paintings. using even a mobile phone camera.

< < Contents 28

DSLRs or compact systems that allow
interchangeable lenses are generally more
versatile than other options, albeit at a higher
price. If you are using a point-and-shoot camera,
try to get one with the highest quality lens that
you can afford.

For photogrammetric work, it is generally best to

avoid using image stabilisation (IS) or vibration
reduction (VR) functions on lenses or cameras
that offer this capability. Although the images
produced will appear sharper, systems of this sort
generally work by moving either the image sensor Figure 28
itself or an optical element group in the lens at or Example of a colour calibration chart in use.
immediately prior to the point of image capture,
both of which result in a slightly offset principal
point, although this offset can be estimated
during the SfM process. for example, 12mm (35mm equivalent), but their
use should generally be avoided, at least in part
The best results are obtained using fixed focal because the resolution can vary considerably
length lenses, especially if these have been across the image and radial distortion is high.
calibrated. Wide-angle lenses (for example around Zoom lenses can be employed, and there will be
28mm) can be very useful for capturing as much situations when their use is unavoidable. Provided
of the subject as possible without introducing too the approximate focal length values are written to
much distortion, reducing the number of images the image EXIF data this is not usually a problem,
required and improving matches between images and even without this information most software
(Figure 29). As can be seen, the longer the focal currently available is extremely good at estimating
length, the narrower the field of view becomes, the focal length. Zoom lenses have the additional
although the magnification factor increases. Some benefit of providing a great deal of flexibility for
software allows the processing of images taken the photographer.
with very wide-angle or fish-eye lenses, up to,

Figure 29
Fields of view achieved by different focal lengths of lens.

< < Contents 29

When imaging very small objects, macro lenses most closely to the colour perception of the
can be used. These usually offer a very shallow human eye (Verhoeven 2010). Verhoeven (2010,
depth of field, so it may be necessary to use a 2016) provides useful and detailed discussions
very small aperture to compensate for this and of this topic. The DNs form a grey-scale image,
allow more of the image to be in focus (see which must then be converted to a colour image
section 3.4.1 Macro lenses). by a process known as demosaicing, in which
the intensity of colour in a particular channel
2.1.8 Image format can be determined for each pixel using the CFA
Almost all digital cameras on the market today are data, and the other values (for those channels
capable of saving images in joint photographic not represented at that particular point) are
experts group (JPEG or JPG) format. Some, interpolated from those around it.
notably those at the lower end of the market, such
as compact cameras, will only output images in RAW files also preserve the whole dynamic range
this format. Other cameras allow the raw data offered by the camera. The dynamic range can
from the sensor to be output in RAW (not an be characterised as the range of luminance that
acronym) format, and some also allow the data to a camera can capture. In most cameras, the
be saved as tagged image file format (TIFF) files. raw image is recorded using 12 or 14 bits (the
bit depth) per channel. Twelve bits offer 4,096
RAW files are minimally processed by the camera, shades per channel; 14 bits offer 16,384 shades
their generation involving simply the conversion per channel. When converted to JPEG, which
of the analogue information gathered by the only offers 8 bits per channel (256 shades), it
sensor to a digital format with some amplification, clearly cannot transmit all of the information
and thus they constitute the ‘digital negative’. available, so some clipping of the dynamic range
Although all digital cameras record in RAW at the is necessary. A tonal curve is applied that will
moment of capture, in cheaper cameras this RAW typically clip highlights at the expense of retaining
file is converted to JPEG immediately and the raw better detail in the darker areas of the image
information discarded. In this case, the user or (Stylianidis and Remondino 2016). This reduced
automatic settings (for example white balance, dynamic range can result in posterisation in the
sharpening and exposure adjustments) are final images.
applied to the raw data when the file is written,
and a clipped tonal curve is also applied. Once A further consideration with JPEG files is their
the file is written, these changes cannot instability during and after processing. Re-saving
be undone. a JPEG introduces compression errors, and this
is compounded every time the file is re-saved,
The RAW format, in contrast, allows the image resulting in gradual degradation of image quality
to be altered post-capture without affecting the (Hass 2007; Verhoeven 2010). Retaining the RAW
original data. A RAW file contains information files means that the original data can be returned
in three main groups: data numbers (DNs), to at any point and re-processed with no loss of
describing the intensity of signal received by quality. Furthermore, nearly all digital cameras
each photo-diode (pixel) on the sensor, the use different quantisation tables when writing in-
configuration of the colour filter array (CFA) camera JPEG files, so there is no real equivalence
overlaid on it, and metadata. The CFA is a colour between the same settings on different cameras.
pattern filter overlain on the sensor that limits For example a Nikon ‘fine’ JPEG setting is not the
the spectral components gathered by each same as a Canon or a Sony setting (Hass 2007).
photo-diode to (usually) red, green or blue; the It is evident, when comparing in-camera JPEG
most common is known as a Bayer array and files from different manufactures with the RAW
favours the green channel over the red and blue equivalents, that some are much more aggressive
channels because this arrangement corresponds than others, whatever the setting chosen.

< < Contents 30

Although RAW files from different manufacturers As the flying time required for an SUA is not
contain the same basic information they vary usually very long at most archaeological sites,
widely in format. This is sometimes seen as a the second option is preferable because it has
drawback, but in practice many image-processing no effect on the GSD, does not adversely affect
programs are capable of handling raw image data image quality and involves little inconvenience
from a wide variety of cameras. However, it can despite the extra flight time. There is usually no
be a disadvantage for archiving. While Historic negative impact if a rotary SUA is used because
England advocates the retention of RAW files, the flying speed can be slow enough for the
these are converted to uncompressed TIFF files camera to write the files.
for archival deposition, largely because the TIFF
format is perceived to be archivally ‘stable’, is If you are using a camera that will only output
supported across most platforms (for example images in JPEG format, it is advisable to use
Windows, Macintosh and UNIX) and can, with settings that yield the largest file size and the
careful processing, represent a minimal loss of lowest compression ratio at capture, and to
information from the RAW camera output. convert the files to TIFF immediately after
download. There is little utility in using in-
In summary, RAW files contain (and retain) camera generated TIFF files if this is an option,
more information and have better compression because that results in the loss of the original
(file size) than TIFF files, and also avoid the RAW information and much larger files. In an
compression artefacts often found in JPEG ideal situation, both RAW and JPEG files should
files. However, there are some practical caveats be recorded; this allows rapid assessment of
to consider when advocating the use of RAW the suitability of the images on-site, and the
photography. RAW files are usually considerably identification of any omissions or potential
larger than in-camera generated JPEG images, problems, before processing the RAW files when
and consequently take longer to write. In some back in the office. Despite all of the above,
circumstances, for example when using a high- however, it should be noted that much of
resolution camera on a fixed-wing SUA, the the more popular software used for SfM–MVS
required time interval for taking shots with processing is able to cope with JPEG files and
sufficient overlap is shorter than the time is often optimised to some degree for them,
needed by the camera to write each file between given the common requirement for their use by
exposures. This has to be compensated for by fixed-wing SUA.
using one or more of the following:
2.1.9 Multispectral Imagery
„ increasing the flight height to ensure Images derived from sensors operating outside
overlap with fewer images, although this that part of the electro-magnetic (EM) spectrum
will have a detrimental effect on GSD and (Figure 30) that is visible to the human eye
may, subject to Civil Aviation Authority (CAA) can be processed using a normal SfM–MVS
regulations in the UK, not be possible photogrammetric workflow and have a wide range
of well-documented archaeological applications.
„ flying the area more than once to obtain They can be used in conjunction with products
complete coverage with a sufficient interval derived from ‘normal’ red, green, blue channel
between exposures to ensure that the files (RGB) imagery in a number of useful ways, for
can be written example comparing and analysing near infra­
red (NIR)-derived ortho-images with their RGB
„ decreasing the focal length of the lens used counterparts and by combining the channels from
to increase the coverage with each image both sets of imagery to generate different outputs.

„ changing the format from RAW to

in-camera JPEG.

< < Contents 31

Figure 30
The electromagnetic spectrum.

NIR aerial photography has been usefully and

comprehensively discussed by Verhoeven (2007,
2008, 2012) and has been evaluated on many
sites in the UK (Dawson and Winterbottom
2003; Powlesland et al 1997). NIR photography’s
popularity derives in part from the relative
cheapness and ease with which the necessary
equipment can be obtained, as it often involves
modification of an existing camera rather than the
purchase of dedicated equipment. NIR imagery
can be used on its own (usually after histogram
stretching and contrast enhancement) or
combined in various ways with the red, green or
blue channels from visible spectrum imagery for
estimating, for example, the normalised difference
vegetation index (NDVI; originally developed for
use with Landsat imagery in the 1970s) or fraction
of absorbed photosynthetically active radiation
(FAPAR), both measures of vegetation health
(there are many others). Examples of such imagery
are shown in Figure 31. In simple terms, diseased
or stressed vegetation reflects less light in the
Figure 31 NIR spectrum than healthy vegetation, and thus
Unprocessed NIR (top) and RGB imagery (middle), appears darker in NIR imagery.
with derived NDVI image (bottom).

< < Contents 32

Crop marks typically occur because sub-surface
features either aid or hinder crop growth by
increasing or reducing moisture availability and
root penetration (Figure 32). They tend to be more
apparent when crops are under stress from lack
of moisture, which emphasises the differences
in water availability throughout a crop. Buried
ditches and trenches tend to retain moisture and
allow deeper root penetration, leading to better
crop growth, while walls tend to retain less water,
leading to shallower root penetration, in turn
leading to more stressed crops. In the visible part Figure 32
of the EM spectrum, the differences observed vary Sub-surface features and their effect on overlying crops.
depending on vegetation type, and are principally
showing the relative concentrations of different
plant pigments (Tucker and Garrett 1977). Images
taken with NIR cameras detect different variations, NIR cameras (typically 640×480 or 320×240 pixels,
typically the number and arrangement of air depending on price), and therefore the images are
spaces in the leaf structure and moisture content. processed separately and formal ground control is
RGB images therefore only show differences that usually required to register it with imagery derived
are perceptible in the visible spectrum, whereas from other sources. Some photogrammetric
NIR imagery, which provides a better reflection of packages may not be able to process imagery
vegetation health, can show features that would successfully at this low resolution.
be otherwise undetectable.
2.1.10 Image enhancement
Although unmodified cameras with NIR filters In general, image enhancement should be avoided
applied to the lens can be used for terrestrial and any functions that change the relative values
NIR photography, the long exposure times of the pixel structures in an inconsistent manner
required to achieve good results preclude should not be used when pre-processing images
their use in aerial photography. For aerial NIR for photogrammetric purposes. Often, those
photography, DSLRs that have been modified by methods that produce more visually pleasing
removing the hot mirror and replacing it with an images, by increasing apparent clarity for
NIR pass filter typically capture infra-red radiation example, in fact adversely affect the image at the
in the 0.7µm to 1.4µm range (the visible part of pixel level and can introduce artefacts.
the EM spectrum is between circa 0.38µm and
0.7µm). NIR images are also useful because the Adjustment of the brightness/contrast (performed
longer wavelengths in this part of the spectrum during the conversion of an image from raw to
are less subject to atmospheric scattering, thus 8-bit) is sometimes useful, as is a minimal use of
significantly reducing the effects of haze in unsharp masking. Neither of these techniques
high-altitude or oblique aerial photography should cause information loss or the introduction
(Verhoeven 2012). of artefacts (Chandler 2010) but do require
prudent use and the results should be checked.
Further up the EM spectrum, thermal imagery Histogram equalisation, which can be used for
can also be used in conjunction with RGB manual photogrammetric digitisation, should
photography to provide additional information, generally not be used for processes that involve
and has proved useful for detecting features in automated image matching: the software can
areas where dense vegetation precludes the use of distinguish slight variations in pixel values
RGB photography (Brumana et al 2013). Thermal without having to apply the exaggeration
imaging sensors are usually very low resolution necessary to make them visible to the human eye
compared with those in most RGB or modified (Chandler 2010).

< < Contents 33

2.2 Image arrangement Forward overlaps for vertical imagery usually
need to be at least 60 per cent, with lateral
Image arrangement for all photogrammetric overlap between flight lines (or side lap) between
work is underpinned by the fact that each point 15 per cent and 40 per cent or higher (often
to be measured/recorded should be intersected 40–60 per cent in SUA imagery), depending on
by at least two rays (see section 1.2 The chief flight height (Figure 34). The Royal Institution of
ray and principles of intersection), although in Chartered Surveyors (RICS) (2010) provides useful
practice intersections from many more images information on this topic.
are desirable. As the recent increase in the use of
photogrammetric data in archaeology has focused A typical image arrangement from a fixed-wing
(at least in part) on software that employs SfM SUA on a pre-planned flight path is shown in
techniques, image capture strategies for these Figure 35. All the imagery is vertical or near-
are (briefly) discussed and illustrated in Image vertical. Vertical surfaces, such as cliffs or
capture strategies. the sides of buildings, are generally not well
represented, and, if these are required, additional
2.2.1 Image capture strategies oblique photography will be necessary. The
2.2.1.1 Aerial degree of overlap achieved by these images is
For aerial shoots, the types of images captured shown in Figure 36.
are to some extent determined by the platform
employed. Fixed-wing SUA platforms will typically A typical arrangement of images of an
capture nadir (vertical), or near-nadir, imagery, archaeological site taken from a manned aircraft
whereas rotary platforms and manned aircraft is shown in Figure 37. All the imagery is oblique.
open up the possibility of oblique image capture In this image configuration, the overlap between
(Figure 33). images is very high and approaches 100 per cent,

Figure 33
Different classes of aerial image.

< < Contents 34

Figure 34
Forward overlap in vertical aerial imagery.

Figure 35 Figure 36
Typical arrangement of vertical images from a The overlap achieved by the imagery.
fixed-wing SUA.

< < Contents 35

Figure 37 (top) Figure 38 (above)
Typical arrangement of images of an archaeological The overlap achieved by the imagery.
site taken from a manned aircraft.

< < Contents 36

depending on the intervals between shots.
Vertical surfaces, such as the sides of buildings, Figure 39 (top left)
are well represented. The degree of overlap Imagery that was supposed to be vertical taken at
Roche Abbey, South Yorkshire.
achieved by these images is shown in Figure 38.
Figure 40 (top right)
Figures 39–41 show the processing of SUA A different image capture strategy could have
ameliorated the problem. The grey arrows indicate the
imagery taken at Roche Abbey, South Yorkshire.
actual flight lines; the green arrow shows a flight path
The processing revealed that the camera had not giving better coverage.
been pointed directly downwards but at an angle
in every shot (Figure 39). Had the images been Figure 41 (above)
taken with the camera angled and moving in a As a result one side of the buildings is missing data.
sequential pattern (Figure 40), the gaps would not
have been significant. However, as all the images

< < Contents 37

 were taken with the camera facing one way up the
site, there was a considerable amount of missing
data on one side of the building elevations
(Figure 41).

2.2.1.2 Terrestrial
General image capture strategies commonly used
for terrestrial shoots are shown in Figure 42. An
image arrangement for the recording of a building
elevation is shown in Figure 43. The lowest
images were taken from a tripod on the ground;
those from higher up were taken using a camera
mounted on a 9m mast at two different heights.
Although a plan view showing the roof only, the
overlaps for the image arrangement shown in
Figures 42 and 43 are illustrated in Figure 44
and are indicative of the overlaps seen on the
building elevations.

Figure 42 (above)
Image capture strategies for terrestrial photography.
Adapted from Agisoft user manual (2017) Reproduced
with permission.

Figure 43 (below left)

Typical arrangement of imagery taken for recording
a building elevation.

Figure 44 (below right)

The overlap achieved by the imagery shown.

< < Contents 38

Figure 45 Figure 46
A building elevation photographed under very The illustration based on the imagery.
compromised circumstances.

Another building elevation is shown in Figure 45. Figure 48 shows the modern re-processing of
In this instance a mast could not be used to gain a series of images taken in a traditional stereo
higher level imagery because of scaffolding, and photogrammetric setup with an analogue camera
the stand-off distance was compromised because in 1997 at Chatham dockyard, Kent. The images
the street was very narrow. This led to gaps in the were taken to provide a series of stereo models
data, notably on the upper edges of window sills that were then combined to form a single ortho­
and above most projections on the facade. As the image. The SfM–MVS approach allowed the re­
required product was a line drawing produced processing of the entire set of imagery at once.
in CAD rather than an ortho-image (Figure 46),
these gaps could be tolerated and the additional For a piece of complex 3D geometry, such as the
information for roof lines and chimneys was carved front of the sarcophagus shown in Figure
infilled using a TST from some distance away 49, additional imagery has to be taken to infill
and very obliquely. possible gaps. As well as a series of images taken
perpendicular to the face of the object, images
The image arrangement for a stone cross is shown were captured looking both down and up, both
in Figure 47. In order to capture the top of the runs overlapping considerably (Figure 49).
subject, a camera mounted on a mast was used.

< < Contents 39

 Figure 47 (left)
Image arrangement used when recording a medieval
stone cross.

Figure 48 (below)
Stereo cover arrangement with an analogue camera
taken before SfM methods were available.

Figure 49
Image arrangement for a relatively geometrically
complex carved object.

< < Contents 40

2.2.1.3 Small objects
In this example, the partial skull of a small dog
excavated from a Roman context at Raunds,
Northamptonshire, was recorded. The skull,
which was extremely delicate, was photographed
using a static camera with the skull placed on a
turntable, which was rotated slightly between
each exposure. Images were taken at three height
intervals as the object was rotated (Figure 50). The
skull was then turned over and imaging repeated
(Figure 51). The two models were aligned using
common points from both models, giving the
composite image arrangement shown in Figure
52. Each of the dense point clouds could then
be cleaned up to remove extraneous points
associated with the base on which they were
placed and the callipers used as scale bars, and a
single unified point cloud produced from which a
mesh could be derived. With this type of project,
achieving a high degree of overlap between the
image sets of both sides of the object is very
important, otherwise uniting the halves of the
model successfully can be extremely difficult. This
type of project is also easier if the subject has
highly textured surfaces (with clearly definable
common points) if control markers cannot be
affixed to the object, as in the example illustrated.

2.2.2 The 3×3 rules

The original 3×3 rules, outlined here (Waldhäusl
and Ogleby 1994), were published as a check list
for the simple photogrammetric documentation
of architecture. Although SfM photogrammetric
techniques have rendered some of the original Figure 50 (top)
provisos redundant (such as keeping the inner Images taken of the upper half of a small dog skull.
orientation of the camera constant), they Figure 51 (middle)
nonetheless remain a very useful set of Images taken of the lower half.
reminders for a methodical approach that will Figure 52 (bottom)
yield great benefits. Composite image arrangement after alignment of the
two halves.

The three geometric rules cover:

„ control The three camera rules cover:

„ wide-area stereo image cover „ camera properties

„ detailed stereo image cover „ camera calibration

„ image exposure

< < Contents 41

The three procedural rules cover: WGS84, may be preferable, whereas for building
survey an arbitrary site coordinate system is often
„ recording control and image layout used. It is important that your software is able to
handle projected coordinate systems correctly
„ metadata when you import control data.

„ archive Photogrammetric control performs the same

function as control in many other forms of
The 3×3 rules have been updated to take account surveying: without it there is no check on internal
of more recent photogrammetric developments errors, which may propagate throughout the
(TheoLt 2010). model, and although things may ‘look right’
there is no guarantee that they are; a lack of
control removes the facility to check for, quantify
2.3 Control and mediate error. Control points are used not
only to locate data spatially, but also to scale
This section considers the use of externally and orientate the data, for optimisation of the
measured control points for the purpose of automatic image alignment and the reduction of
model refinement, scaling, orientation and non-linear errors in the model, and for checking
checking. Clearly, the SfM process itself involves the accuracy of the reconstruction.
the identification of large numbers of tie points
between images, and these constitute an internal Control may not be required in some situations,
control network of sorts in their own right. If for example for work undertaken with only
you are using a package without the facility to visualisation in mind or with low metric
introduce external control measurements during requirements. For accurate survey work, or where
the bundle adjustment, the tie points are the measurements may need to be taken from the
only correspondences that are used, without model, control is essential, especially if the results
a measured ‘real-world’ spatial component. are to be integrated with the products from other
However, the main concern of this section is survey methods. Subjects that are ‘full 3D’ (for
measured, coordinated points on or around the example a statue) will tend to perform better (as
subject of interest. models with no control) than those that are 2.5D
(for example most aerial subjects), as the imagery
Control for photogrammetry usually comprises a should ‘close’ all round the subject, reducing the
set of clear and unambiguous points that appear chances of cumulative error propagation.
in the images and for which the coordinate
positions are known. For aerial surveys, camera Ideally, control should be an integral part of a
positions are also often recorded [using an on­ project from start to finish. Redundancy in
board GNSS and inertial navigation system (INS), your control network is useful: you should
for example] and written to the image files as EXIF attempt to have more control points than you
data; these can be used for approximate scaling need, so that some of them can be used for
and orientation of an otherwise unconstrained optimising alignment, and scaling and
model if no ground control is available. orientating the project, and others can be used
afterwards as check points to verify the accuracy
Control points can be in any reference system of the reconstruction.
or coordinate frame, depending on the source
data. The Ordnance Survey National Grid (OSNG) It is beyond the scope of this guidance to
is often used in the UK for applications where deal with the theory and practice of control
spatial location relative to the national mapping network design. There are, however, a few basic
framework is important. For other outputs, considerations that can be applied successfully
reference to a global coordinate frame, such as to almost any project.

< < Contents 42

2.3.1 General considerations
2.3.1.1 Number of control points
Photogrammetric models are, on their own, scale
free. For scaling, a minimum of two control points
is needed. These need not be 3D coordinates
but may be points identified at the ends of a line
of known length measured between two points
visible in the images, for example a scale bar.
This information can be used to scale, but not
constrain the formation of, the model, which
remains in an arbitrary coordinate frame.

For scaling and orientation, a minimum of Figure 53

three points is required, two of which must be Additional control points are required in areas of
topographically more variable terrain.
3D coordinates (that is x,y,z values) and one of
which need only be 1D (x,y or z ); they must all
be visible in at least two images, but in practice
should be visible in many more. Most software will
in fact require three 3D coordinate values. With 2.3.1.2 Distribution of control points
this number of control points, the model can be You should try to keep control points evenly
scaled, positioned and orientated relative to a distributed across the area of the model rather
coordinate frame, and thus achieve a degree of than grouped in a particular area or in a straight
absolute accuracy, but the relative accuracy of the line (Figure 54). For aerial recording, place
model will remain unaffected (the control points points near, rather than at, the edges of your
are used for scaling and orientation but do not area of interest, to ensure that the edges receive
contribute to adjustments in the formation of enough image coverage. If you do need to place
the model). the control points right at the edges of the
study area, try to extend the flight area slightly
If you wish to use control points to refine the to ensure sufficient coverage. One or more
estimated image alignment and hence the additional points placed centrally within the area
accuracy of the reconstruction, you will need is often sufficient if the topography is relatively
more than three 3D points. It is therefore even; if there is considerable variation, placing
advisable to use more than the minimum number points near, for example, the tops and bases of
of points, but the number of control points significant topographic features is helpful. The
required varies to some extent with the complexity same principle holds true for architectural and
of the subject. For aerial surveys, for example, other recording: distribute controls as evenly as
areas with relatively little topographic variation possible across the subject, and pay attention
may require as few as 6–10 control points, to significant projections or recessions. Most
whereas the accurate reconstruction of more photogrammetric software that deals with
topographically variable terrain will benefit from external control has extensive documentation that
additional points (Figure 53). Similarly in building discusses control points and their use, and it is
survey, fewer control points are required for well worth reading this before starting a project.
accurate reconstructions of relatively flat facades
than for more geometrically complex subjects. 2.3.1.3 Accuracy and control points
However, adding ever more control points to the It is useful to distinguish between relative
adjustment yields diminishing returns in terms accuracy (the accuracy of the photogrammetric
of increased accuracy. In all cases, additional reconstruction itself) and absolute accuracy (the
measurements that can be used later as check accuracy with which the model is placed within a
points (not to refine the alignment but to verify coordinate frame). The number of control points
the accuracy of the reconstruction) are advisable. used can contribute to absolute accuracy only

< < Contents 43

Figure 54
Control point distribution.

Control markers for aerial survey (for example centres of manhole/drain covers)
In order to mark ground control points can be used, but temporary targets, such as
(GCPs) accurately in images, they must be of spray-painted crosses using dispersible, non­
sufficient size to be clearly visible. Thus for toxic paint, may be necessary, especially in
aerial photography, the target size should be areas where hard detail is lacking. A wide
approximately 5–10 times the ground sample variety of aerial photography targets is
distance (GSD) of the survey, both to enable available from most survey suppliers, and
it to be clearly visible and for the point to be some examples are shown in Figure 55. It is not
marked accurately; for example, for an aerial advisable to use building corners or points very
survey with a GSD of 40mm, targets between close to tree cover, because global navigation
20 and 400mm across (or larger) should be satellite systems (GNSS) accuracy is often
used. There are many choices regarding the compromised in such areas.
type of marker used in the field, but they must
be of sufficient size for the measured centres
to be clearly visible in the images. In some Figure 55
cases well-defined ground features Sample aerial photography targets.

< < Contents 44

(placing an otherwise unconstrained model in Because of these factors, check points are the
a coordinate frame) or to both absolute and best method for assessing the metric quality of
relative accuracy (constraining and refining model the outputs. The relative accuracy for SUA surveys
creation as well as scaling and orientating it). after processing is approximately 1–2 pixels in
The accuracy of GCPs is mainly dependent on plan and 2–3 pixels in height (TSA 2013). If you
(a) the positional accuracy of the method used use survey-grade GNSS to locate a suitable
to derive them and (b) the accuracy of their number of GCPs across the survey area, the GSD
identification and placement in the images. If can be used to estimate the accuracy of the
only camera location data is used, no image survey, thus with a GSD of 30mm the accuracy can
placement is necessary as the recorded camera be estimated at 3–60mm in plan and 60–90mm in
positions themselves provide the control. Control height (TSA 2013).
should be measured with a degree of accuracy
appropriate for the general requirements of the 2.3.2 Sources of control data
project, and need not be of a higher accuracy than Control data can be derived from a number of
is necessary. Survey-grade GNSS coordinates with sources. The method chosen will be dictated by
a positional accuracy of 1–30mm will not usually the accuracy requirements of the project.
be economic for a project where metre or broader
accuracy is all that is needed, and alternative 2.3.2.1 GNSS (aircraft/SUA on-board)
control sources can be more appropriate. The typical accuracy of the on-camera GNSS units
used by Historic England (the Nikon GP-1, with
The accuracy of an entire project is subject to a a manufacturer’s quoted accuracy of 10m RMSE
large number of variables and should match the horizontally) is commensurate with the accuracy
initial specifications for the accuracy required. of most navigation-grade GNSS equipment. The
These variables include, but are not limited to: typical accuracy of in-aircraft GNSS units for
light aircraft (for example the Garmin GPS map
„ image configuration (for example flight 496) is 15m RMSE 95 per cent, down to <3m with
plan/completeness of coverage, flight differential global positioning system (DGPS)
height/stand-off) corrections (Garmin 2007). Most SUA have GNSS
receivers with a similar performance, although
„ sensor quality/resolution some are available that use a DGPS solution
to improve locational accuracy significantly.
„ image quality GNSS values are recorded at the point of image
capture and embedded in the EXIF data of the
„ ambient conditions (for example weather, images. Developments to increase accuracy in
lighting) archaeological prospection include combining a
GNSS receiver and an INS for estimating exterior
„ accuracy of control method chosen (for orientation at the point of image capture from
example GNSS, TST, scale bars) and manned aircraft, to give an image position
marking on images to approximately 2.5m and orientation to
approximately 2° (Verhoeven et al 2013).
„ distribution and number of control points
Such data can be used by most photogrammetric
„ the relative skill of the surveyor software for initial relative estimations of image
position and, in cases where high accuracy is
„ the processing techniques to be used not a priority, can provide all the locational and
and software options chosen. control information necessary for a project. In
most other aerial work the data will be used for
the initial orientation only, and will be replaced as
control points by higher accuracy ground-based

< < Contents 45

 GSD and mapping scale A useful tool for calculating GSD if the other
The ground sample distance (GSD) is the factors are known can be accessed via the
distance on the ground that is represented by Pix4D website.
the distance between adjacent pixel centres
in an image (so in an image with a GSD of Reducing the GSD, assuming the same camera
50mm, each pixel will represent 50mm×50mm = system is used, requires more flight lines
250mm 2), and, in digital cameras, is a function (because of lower flying altitude) and will
of the pixel dimensions of the sensor array, the produce more images, and hence require a
focal length of the camera and the flying height longer processing time. It is recommended
(Neumann 2009). It represents the spatial that, if procuring aerial photography from a
resolution of an image, so it is only applicable third-party contractor, the GSD requirement
(across a single image) to imagery taken should indicate a target GSD and a maximum
perpendicular to the subject, such as vertical permitted GSD in order to take account of
aerial imagery, or in an ortho-image. In oblique terrain variation. Features that are smaller than
photographs it will vary across the image as the the GSD will not normally be discernible on the
distance to the subject also varies (with pixels imagery, although sub-pixel interpolation is
closer to the camera having a smaller GSD than possible with multiple images.
those further away). It will also vary in vertical
aerial photographs according to variations in In general, a GSD of 40mm (equivalent to an
terrain (higher points on the ground, such as image scale of 1:3 000, although dependant on
the tops of hills, will be closer to the camera the resolution of the image) is commensurate
than lower points, such as the bases of valleys). with a mapping scale of approximately 1:500
(with a horizontal RMSE of ±0.1m and a
GSD may also be calculated for an output vertical RMSE of ±0.05m; RICS 2010). The Royal
ortho-image generated from a number of Institution of Chartered Surveyors (RICS) lists
oblique inputs, and most photogrammetric commonly used topographic mapping scales
software will give an indication of the GSD of and their GSD requirements (RICS 2010), and
the product and allow it to be changed. It is a The Survey Association (TSA) has some useful
useful way of indicating the level of detail that guidance notes covering this topic (TSA 2013,
can be reasonably expected from the inputs, 2015).
based on their resolution.

coordinates. When using locational EXIF tags, ground-based GNSS coordinates using survey-
which are usually written to image files in the grade equipment varies according to a number
WGS84 coordinate frame, it is worth remembering of factors but should broadly be within 10–40mm
that the height reported is the height above the under normal conditions (Where on Earth Are
ellipsoid rather than the altitude of the aircraft We?; accuracies for other grades of survey
relative to the ground. equipment are also given). This level of accuracy
is clearly commensurate with the resolution of
2.3.2.2 GNSS (terrestrial) imagery acquired by SUA and the accuracy of the
Survey-grade GNSS is a common method of anticipated outputs. It is worth keeping an eye on
locating GCPs for aerial survey work, whether the reported accuracy of your GCPs, as it will vary
from manned aircraft or SUA. The accuracy of with time and location.

< < Contents 46

In many cases the GCP coordinate positions reflectorless at a distance of around 50m or less,
derived from ground-based GNSS measurement which is the usual range for these applications.
will be relative to a national coordinate frame If your stand-off is much greater you will not be
(for example the British national grid) rather than able to see the centre of a small target properly,
the WGS84 coordinates derived from the image and you need to remember that the laser dot
EXIF tags. If this is the case, whatever software is size of the TST is variable with range because of
being used, remember to assign the appropriate beam spreading (for example, for a Leica TS15i
coordinate system, so that the later products can it is circa 7×10mm at 30m and 8×20mm at 50m).
be located correctly relative to data derived from Targets can range from small but unambiguous
other sources. Assigning incorrect projections by points of detail (for example the screws on a
mistake may not be apparent immediately but can light fitting, ventilation covers, and corners of
lead to significant errors in placement. windows in building survey) to small adhesive
stickers and traditional photogrammetric butterfly
2.3.2.3 TST targets (Figure 56). If the subject is being laser
A TST is often used for control work in building scanned simultaneously, scanner targets can be
survey or archaeological excavation. It will incorporated into the control network, although
typically provide points that are accurate to the scanner data itself will provide sufficient
around 2mm +/- 2 parts per million (ppm) control on its own in most circumstances,

Figure 56
Sample targets for terrestrial survey work.

< < Contents 47

especially if combined with imaging (see Laser TST as they are measured and referenced to
scanners/lidar). Spherical scanner targets are the defined control points (Figure 57). This can
clearly not usable in this way. ameliorate any ambiguity arising from inadequate
site notes, but bear in mind that photographs
If you are using points of detail, which is the least taken with survey instruments often perform
intrusive method and can be the only option poorly if they are pointing towards a light source
available if targets cannot be placed on surfaces, or in low light conditions (Figure 58), so additional
it is vital you keep good site notes showing narrative photography and notes should be taken.
the points so that they can be identified easily As with aerial control, you need to be able to
and accurately after fieldwork is complete. It identify the target unambiguously in the images,
is surprising how quickly information can be and to be able to define the measured centre.
forgotten, and surveys potentially compromised,
without good notes. Using an imaging TST, if 2.3.2.4 Laser scanners/lidar
available, is extremely helpful in this regard, as Data from laser scanners can provide excellent
images of control points can be taken with the control for terrestrial photogrammetric projects.
If the density of the scan data is sufficient,
coordinates can be derived directly from the
point cloud and these values input for the
corresponding positions on the images. Scanner
targets (often used for registering point clouds
from different setups), if present in the imagery,
can also be used: these are often detected
automatically by the scanner and scanned at a
higher resolution, helping ensure sufficient data
density in these areas.

If data from photogrammetry is to be integrated

successfully with scanner data to form a
composite product, or the scanner data is
to be used as a model for texturing with the
photogrammetric images, then the scanner data
has to be used as the control basis in order to
provide sufficiently accurate matching. Alternative
methods, such as using target values measured
independently by a TST, do not usually provide
the best fit. Laser scanner data works well as
the control for a range of size of subjects, from
building surveys to smaller objects. However,
it is far easier to determine the position of a
control point from laser scan data if the scanner is
recording images as well as range data (although
in many scanners there is a time penalty for
this); some software (such as RealityCapture,
can combine photogrammetric and scanner data
Figure 57 (top) but the imaging has to be taken on the scanner
Use of an imaging TST to aid site notes. for successful matching of the two data sets.
Figure 58 (above) Bentley’s ContextCapture is also useful in this
Example of a poor image from an imaging TST regard, and does not require colourised scans
in challenging lighting conditions.
for matching.

< < Contents 48

For aerial photogrammetry, lidar can be used to
provide control data. At typical lidar resolutions
[for example, the Environment Agency (EA) data
varies between 0.5m and 2m for most of the UK],
the accuracy will not be the same as that derived
by survey-grade GNSS but will be significantly
better than that from global positioning system
(GPS) EXIF tags alone. Smaller, lightweight lidar
units that can be carried on a SUA typically scan
at much higher resolutions than those in manned
aircraft (largely because of their closer proximity
to their target) and have the potential to provide Figure 59
much more accurate control if a photogrammetric Example of scale bar placement as used on the

NADRAP project.

survey is also required (for example for a high-

England’s Rock Art (ERA)

resolution ortho-image). However, there are

relatively few such units flying commercially in
the UK at the time of writing.

2.3.2.5 Scale bars

For smaller subjects, scale bars can be very useful
where formal control is not possible, and are
also useful as a checking mechanism even if a
control network is available. They can be used for
scaling but not orientation, although right-angled
scales can be used to define the x and y axes of
an arbitrary coordinate system. Alternatively, set
squares that are graduated on both perpendicular
axes (preferably in the same units) can also be
used for the same purpose.

Some successful photogrammetric community

projects, such as the Northumberland and
Durham Rock Art Project (NADRAP), have used
scale bar arrangements successfully (Figure 59).
Scale bars are accessible if survey equipment
cannot be afforded, are easy to set up, and
provide sufficient accuracy if absolute orientation
is not required. However, one potential issue with
set squares and scale bars is that they can occupy
valuable space on the image and hence reduce
the overall resolution of the image set. It is often
useful to take both overall shots including the Figure 60 (top)
scale bars and more detailed images focusing Gridded matte drafting film to aid definition of scale
and orientation.
closer in without the scale bars showing or
encroaching on the subject. Figure 61 (above)
Coded targets around the object.
When imaging smaller objects, it can be useful
to have a sheet of gridded matte drafting film or
graph paper underneath the subject, provided it is

< < Contents 49

kept flat. Points on the film or paper can then be There are products available in the UK that can
used to provide scale and orientation (Figure 60). be used if ground control is otherwise unavailable
and the required accuracy is low, for example
2.3.2.6 Coded targets OS Terrain 5, which has a 5m grid of spot
Many photogrammetric software packages heights available as well as contours, although
allow the use of coded targets to provide background mapping will be required.
arbitrary control. Such targets can be
recognised automatically by the software as As a last resort, if geolocation, however
control points to help align images. They can be approximate, is required, there are many web
printed out and re-used from project to project, map services available, including free services
and are extremely useful when reconstructing such as Bing maps and Google maps , from which
objects digitally. Examples of coded targets are very coarse coordinate values can be obtained.
shown in Figure 61. Such targets can also be used Coordinates derived in this way should only be
at fixed spaces to provide scale bars that can used for approximate scaling and orientation, not
be placed around the subject (for example see image alignment optimisation. The resolution of
Cultural Heritage Imaging). the data is not consistent across the globe, but
these services can be appropriate if the desired
Coded targets must be of an appropriate size output is a low-accuracy .kml file for use in an
relative to the subject so that they can be environment like Google Earth.
identified accurately by the software. Their
application in aerial survey is, generally speaking,
impractical because of the dimensions they 2.4 Historical imagery
would have to be produced at.
SfM techniques can be used with historic
2.3.2.7 Other sources photographic data if a few basic conditions are
For some projects, control values may have to met. The quality of the output will be dependent
be used from other sources. For aerial projects, largely on the following factors:
it is relatively straightforward to obtain x, y
coordinates from a variety of mapping sources, „ sufficient overlap/completeness of coverage
the accuracy of these depending on the scale of
mapping used. It is not quite so easy to obtain „ sharpness and focus
z (height) values of suitable accuracy, and
most photogrammetric software will require 3D „ print distortion (if prints rather than
coordinates for control values. negatives are used)

If you have access to national mapping agency „ images are not cropped
data, such as Ordnance Survey in the UK, you
can use features visible in aerial photographs Research into the metric performance of historical
that have level information available from stereo aerial photographs (Papworth 2015; see
Ordnance Survey maps, for example spot heights case study 1) has demonstrated their potential
on roads or manhole covers. These will only be for assessing and quantifying change over time.
approximate because they do not often coincide Imagery taken with ‘traditional’ photogrammetric
with features visible on the roads themselves, outputs in mind can also be used and re­
so the accuracies are limited to between 0.5m processed, including military reconnaissance
and 3m. The distribution of such points around photography and national mapping agency
a project area may be suboptimal for control but data. One of the major advantages of stereo
may be sufficient for orientation and scaling if no photogrammetric imagery has always been that
other data is available. it does not require immediate processing: once
the images have been captured it is not always
necessary to go the expense of processing them,

< < Contents 50

Figure 62
Historical imagery (inset) producing a poor model due
to lack of clarity.

and they provide an ante-disaster record that can taken with photogrammetric processing in
be used in the event of an emergency. Imagery mind, but it provides a useful cautionary tale
acquired at Windsor Castle, Berkshire, before in the use of such imagery (Figure 62). One
the fire of 1992, for example, was subsequently problem is that the images are not sharp and
processed to inform the reconstruction and re­ in focus; while they show the general form of
instatement of much of the damaged interiors. the landscape under good lighting conditions,
they are noisy. Furthermore, they were scanned
With pre-digital imagery, it is often not possible from prints rather than negatives. The results
to know with certainty which cameras or lenses show that although the general landform has
were used, although this can be established more been extracted, the noise in the inputs has led to
easily for aerial rather than terrestrial images. noisy outputs, with the ground surface exhibiting
Most agencies collecting aerial data on a regular a porridge-like texture, and a large amount of
and systematic basis recorded metadata either extraneous geometry being generated both above
in accompanying log books or around the edges and below the general surface.
of the image frames themselves; if these are
available they should be used. When scans of prints are used (see also
desktop scanners (page 62)), it is essential for
Figure 62 shows an example of a poor-quality the photogrammetric process that the entire
model derived from oblique archive aerial image frame is provided, otherwise the interior
imagery. It is acknowledged that this was never orientation of the camera cannot be estimated.

< < Contents 51

You should always be circumspect if given
imagery that is tightly cropped with no visible
indication of whether the image represents the
entirety of the original or not. Similarly, it is not
always immediately apparent whether a tilt/
shift lens was used; if this is the case, results will
be metrically poor because the estimations of
interior orientation will be incorrect.

When comparing modern and historical data,

there are several points to consider. It is usually
best to compare imagery of similar scales, in
order to avoid inconsistencies deriving primarily
from variations in scale. For example, comparing
historical Ordnance Survey vertical imagery
taken at a flight height of approximately 10,000ft
(3048m) with modern oblique imagery taken from
a height of 1000ft (304.8m) can produce results
that are potentially misleading. The GSD of the
imagery taken from a higher altitude will often
be considerably less than the GSD of the lower
altitude imagery, and comparisons will therefore
be compromised. Such comparisons are not
wholly invalid, but you need to be aware of the
inconsistencies in the inputs and take them into
account when interpreting the results.

Consistency of control is another factor that

has to be borne in mind, and is to some extent
allied with issues of scale. The accuracy with
which control points can be placed on small-
scale images will be less than that in large-scale
images. Additionally, and especially if there has
been considerable variation of the ground surface
over time, choosing consistent control points
can present difficulties. As the control points will
form the basis for the comparison between the
data, they need to be as consistent as possible,
and as many common points should be used as is
practicable. Historical imagery can also be used
for the reconstruction of lost sites, for example as
a result of coastal erosion or wartime destruction.

< < Contents 52

3 Scales and

Applications

One of the distinguishing features of a convergent multi-image SfM photogrammetric

approach is its sheer versatility at a range of scales. The same cameras (and to some
extent lenses) can be used for anything from landscape survey to small objects. The
approaches at varying scales have many similarities, and the common factors are
addressed in General considerations (section 2). This section addresses the special
considerations needed at different scales, focusing on the platforms used to obtain
the images and the characteristics of the subject matter.

3.1 Aerial photogrammetry for Before discussing some of the more

archaeology, landscapes and buildings accessible ways of acquiring and using aerial
photogrammetric data, some distinctions between
There are many potential applications for aerial it and other commonly used data sets need to
photogrammetry in archaeology. It is a well- be made. Lidar data is used extensively in aerial
established discipline in other sectors, with many archaeology (The Light Fantastic: Using Airborne
years of ‘traditional’ photogrammetric processing lidar in Archaeological Survey). While a lidar data
producing mapping for national mapping set is often used as a raster image (in, for example,
agencies, military and industrial applications and a GIS), this image will have been derived, in most
many others worldwide. Archaeological uses of cases, from full-waveform lidar scanner data, in
aerial photogrammetry have, until recent years, which first and last returns can be discriminated
been less prominent, largely because of the cost along with many subdivisions between them.
of image acquisition and the equipment and This allows, for example, the recovery of data
expertise required to process the photographs, representing the ground surface beneath a tree
although single-image rectification using height canopy in wooded areas by filtering the data
displacement correction has been a mainstay of and using last returns to form a digital terrain
the Historic England (English Heritage) national model (DTM). Photogrammetric data is, in these
mapping programme for more than 20 years. The terms, first return only: if the camera cannot see a
advent of affordable photogrammetric software sufficient amount of the ground surface beneath
coupled with more powerful computers and the a canopy, it cannot be modelled because there is
availability of suitable SUA and good-quality no data (there are no pixels in the input images)
digital cameras has seen an explosion in the use representing it. Thus photogrammetry is not
of low-level aerial photogrammetry across all particularly useful in heavily wooded areas. As
geo-information sectors, archaeology not least such, a DSM (as opposed to a DTM) is the usual
among them. product. With regard to this, in areas with few or
relatively well spaced trees, oblique photography
that shows the ground surface under the tree

< < Contents 53

canopies can be extremely useful for filling in any surface ‘beneath’ them will be interpolated
the gaps when combined with vertical images, from the surrounding surface if the holes are filled
and allows the safe removal of canopy-related in (which will happen if, for example, a raster
points without detracting from the quality of the surface is exported to GIS).
derived DTM.
One of the significant advantages of
Similarly, in areas under agricultural cultivation, photogrammetrically acquired data over lidar
the DSM derived by photogrammetry will is resolution. Lidar is available for much of the
represent the tops of the crops planted (first UK at a nominal resolution of 2m (an average
return) rather than the ground surface below (last post spacing of one return every 2m 2), with
return): as already mentioned, photogrammetry some areas covered at 1m, 0.5m and 0.25m
provides a first-return only output so, unless the resolutions. While this provides a considerable
crop is sufficiently sparse, the ground surface will amount of information regarding larger features,
remain invisible. This is not necessarily a major such as those typically identified during aerial
disadvantage, but it should certainly be borne in prospection, in many cases it does not allow
mind when processing the data, especially when the identification of smaller surface features.
incorporating ground control measurements Photogrammetric data derived from an SUA or a
that have been taken from the surface of the soil manned aircraft will typically have a resolution
in which the crops are growing. Depending on between 0.02m and 0.1m, permitting the
the maturity of the crop, significant differences identification and analysis of much more subtle
between the heights of the GCPs and the derived features, although usually, in the case of an SUA
surface may become apparent. Some filtering over smaller areas and discrete sites, the derived
of photogrammetric point clouds is possible, data sets can be huge and flying times are limited.
however, and removal of noise in the data set A manned aircraft, in contrast, allows many sites
from sparse vegetation can be undertaken, as (20+) to be covered during a single flight. Despite
well as the removal of large areas of tree canopy the differences in resolution, once the data has
and buildings (see case study 3). Care should be been derived the analytical stages are very
taken with other features likely to be visible in the similar to those employed when using a lidar DTM
images: static livestock and vehicles, for example, as the source.
will require manual removal, and once removed
With some aerial platforms you may get parts
of the aircraft in the photographs, particularly if
very large wide-angle lenses are used (Figure 63).
If this is the case, you will have to mask-out the
aircraft from the inputs, if possible, or re-fly the
area. If part of the aircraft is in the same place in
every photograph, some software packages have
mechanisms to cope with this (normally used for
masking fiducial marks and other data commonly
placed around the edges of aerial images). If the
location of the aircraft image varies, the images
will normally require manual editing to mask the
aircraft. Kite-derived photography will commonly
have control lines and the operator visible in
vertical or near-vertical shots, though these are
not as problematic as parts of an aircraft or SUA.

Figure 63 In all cases, overlap is essential. Vertical aerial

Parts of the SUA airframe visible in the photography. photography will typically have a front-to-back
© Skyline Images
overlap of between 60 per cent and 80 per cent,

< < Contents 54

with a side lap between swaths of 40–60 per cent
(see section 2.2 Image arrangement). Oblique
aerial archaeological photography, such as
that typically taken from a light aircraft, should
achieve considerable overlap between images
when orbiting the subject, preferably without
leaving gaps around one side of the subject (so as
near a complete circle as possible). It is generally
best to remove any high oblique ‘scene-setting’
shots when processing (Figure 64) as they will
rarely align properly. The lighting requirements
for images to be processed photogrammetrically
are more forgiving than those usually used
in archaeological aerial photography, which
traditionally uses low-raking sun angles to
emphasise the 3D nature of the features on
the ground (see section 3.4.2 Lighting). For
photogrammetric purposes, images taken on
overcast days with even lighting across the scene
are generally preferable; provided the images
are of good enough quality, the software will be
able to compute the 3D aspects of the site, which
can then be emphasised by other means during
later processing, for example by moving a ‘virtual
sun’ around the surface in a GIS to emphasise
or elucidate features. However, this does not Figure 64
preclude the use of more ‘traditional’ aerial High oblique imagery that is unlikely to process well.
photography for SfM: ortho-images generated
from such inputs are visually pleasing and easier
to interpret, especially to the untrained eye
(Figure 65).

You need to place GCPs before undertaking

photography of subjects with poor textural
variation, for example an apparently featureless
grassed field where no distinctive points can be
discerned, or extend the target area to include
details that will allow both matching and
geolocation. On a windy day, when the grass or
crop is likely to be moving and the surface will
be slightly different in each shot, very low-level
aerial photography will not yield results that are
interpretable without ground control. This is
rarely a problem when using photography from a
manned aircraft (because the flight height almost
always allows the capture of detail that can be Figure 65
used by the software to incorporate matching) or Aerial imagery taken with a low sun angle provides an
easily interpretable ortho-image.
with an SUA, but it becomes more problematic
the closer to the ground you are, when less of the
subject is visible in each photograph. In these

< < Contents 55

 cases, the GCPs can be used to help correct the
orientation of the images as long as sufficient
numbers of them are visible; an alternative
strategy is to increase the flight height to ensure
that more of the surface is visible in each image.

Other problems that are encountered with

aerial imagery typically include those that result
from variations in ambient conditions during
photography and those that result from variations
in ground conditions. An example of changing
ground conditions is shown in Figure 66. This site
was photographed once in the morning, when
some snow cover was present, and again later
the same day, when the snow had melted. In such
cases, matching between the different runs using
all the inputs at once is likely to be poor, so they
need to be processed separately and the dense
point clouds merged later using GCPs. A subset
of the images can be used for producing an
ortho-image or texturing.

Significant changes in ambient light (for example

Figure 66 bright sunlight at the beginning of a run or heavy
Changeable ground conditions during photography. cloud cover towards the end of a run) are often
encountered in the UK, although this usually only
affects longer capture periods. If only a surface
model is required, and the differences in lighting
are not too great, then variably lit image sets
can be used. If an ortho-image is needed then it
is worth waiting for relatively stable conditions
to ensure that the result is consistent across
the entire image, and to make it easier to carry
out the image matching. Some software offers
colour correction when generating textures or
ortho-images from inputs with variable lighting,
although this can significantly increase the time
taken to generate the outputs. It is usually better
to take the images under consistent lighting
Figure 67 conditions in the first place, if conditions allow.
Haze in aerial imagery taken from a manned aircraft.
Haze in aerial photographs can be a problem,
particularly with images taken from higher
altitudes (from a manned aircraft), because it
obscures ground detail and makes the matching
of points on the ground surface more difficult.
Variability in haze can also lead to inconsistency
between shots, even in a single run. Photographs
with significant interference from atmospheric
haze will not process well and should be

< < Contents 56

discarded if possible. If photogrammetry is an successful photogrammetric processing of the
intended product, you should choose conditions outputs. Two Cessna 172s are used by Historic
when the air is relatively clear for photography, England for manned aircraft image acquisition,
and avoid oblique shots as far as possible: haze using Nikon D3X and D810 DSLRs. Typical outputs
becomes more of a problem the more oblique the are sets of vertical, near-vertical and oblique
shot because its effect on more distant subject images for each site on the flight plan, the oblique
matter is more pronounced (Figure 67). images usually requiring circling or arcing around
the site (Figure 68). Many sites are often included
GSD is explained in box 3. For the different scales in a single flight, for reasons of economy, which
of output often associated with topographic or means the images have to be separated into
archaeological landscape survey, the maximum discrete sites before processing by most software.
GSDs accepted by Historic England when Although oblique and highly convergent image
procuring survey from external contractors sets are a long way from those captured for
(Andrews et al 2015) are shown in Table 2. traditional cartographic photogrammetry, they
process very well in SfM–MVS-based packages
because the overlap on the subject matter can
Output scale Maximum GSD be up to 100 per cent. A few input images, if
of sufficiently high quality and with locational
1:100 10mm
EXIF tags, can be processed extremely rapidly
1:200 20mm to produce ortho-images and other products if
absolute accuracy of a high order is not required.
1:500 40mm
If high-quality sensors are used, GSDs of 50mm
are easily achievable. In this case, ground control
Table 2 obtained from lidar data can place the model to
within 1 or 2m with very little additional effort
(Figure 69).

Table 3 shows suitable GSDs (approximate values) There are significant advantages to using
suggested by the RICS (2010). manned aircraft, with or without vertical imaging
capability (from a port in the bottom of the
aircraft through which to point the camera),
Output scale Maximum GSD including the number of sites that can be covered
in a single flight, the wide areas over which data
1:1 250 75mm
can be gathered and the weight and number of
1:2 500 150mm sensors that can be carried. The GSD achievable is
comparable with that from a normal SUA at 100m
Table 3 and more than adequate for most archaeological
survey purposes. With the right camera and
lens combination, manned aircraft photography
can also be extremely effective for aerial city
3.1.1 Platforms modelling. The key disadvantage for smaller
3.1.1.1 Manned aircraft operations is a lack of pilots offering the service
The use of manned aircraft for archaeological for imagery of single sites, in which case the cost
aerial photography is well established, well benefits of SUA in comparison become apparent,
understood and extensively covered in but for Historic England the in-house capability
archaeological and scientific literature. It will provides good value for money given the large
not be dealt with in great detail here, other than number of sites distributed over wide areas that it
to outline the photographic requirements for is required to cover.

< < Contents 57

 is the term that is increasingly being used across
Europe, although other popular terms include
small unmanned aircraft (SUA, used here) and
unmanned aerial vehicle (UAV), as well as the
ubiquitous ‘drone’.

SUA are defined by the CAA in the UK as ‘any

unmanned aircraft, other than a balloon or
a kite, having a mass of not more than 20kg
without its fuel but including any articles or
Figure 68 equipment installed in or attached to the aircraft
Aerial archaeological photography from a manned at the commencement of its flight’, and by the
aircraft – circling configuration.
International Civil Aviation Organisation (ICAO)
as ‘A set of configurable elements consisting
of a remotely piloted aircraft, its associated
remote pilot station(s), the required command
3.1.1.2 SUA and control links and any other system elements
There is a plethora of terms for small aerial as may be required, at any point during flight
vehicles capable of carrying a sensor (in the operation’ (ICAO Circular 328). SUA fall broadly
photogrammetric case usually a camera). into two categories: fixed-wing and rotary. Both
Remotely piloted aircraft systems (RPAS) have advantages and disadvantages.

Figure 69
Coordinates from lidar used as ground control.

< < Contents 58

Fixed-wing SUA tend to give a more ‘traditional’
photogrammetric output, usually shooting
overlapping swaths of vertical or near-vertical
photography in the most efficient manner. They
can typically cover quite large areas in a single
flight (generally about 1km 2, which is the limit
of legal ‘line of sight’ CAA regulations in the
UK) and are thus effective for the surveying and
mapping of landscapes and open sites. Flight
durations are typically up to 45min on a single
battery. With spare batteries, good weather and
good planning, several flights can be undertaken
in a day. For relatively straightforward mapping
cases (open sites with earthworks, for example),
vertical imagery on its own is perfectly adequate,
but often a combination of vertical and oblique
imagery is needed, not only for more complete
representation of vertical elements but also
because oblique images can be usefully employed
to improve the bundle block adjustment,
and hence the accuracy of the reconstruction
(Nocerino et al 2013).

Lightweight fixed-wing SUA normally use lighter

and smaller cameras than those available with
other systems, as they generally have a smaller Figure 70
payload capacity. There are consequences arising Poor images taken from a fixed-wing SUA under windy
conditions: GCPs shown.
from this limitation. Most of the cameras currently
carried by fixed-wing SUA are in the compact
point-and-shoot category and generally output
JPEG files, although different cameras can be
used with some of these SUA. For the reasons generated from them are of sufficient quality for
discussed in section 2.1.2 Resolution and sensor many archaeological survey applications. Fixed-
size, the quality of images from the default wing SUA can generally operate with wind speeds
lightweight fixed-wing SUA cameras is generally up to 65km/h (TSA 2015). Their great advantage
less good, and the settings limitations imposed is speed and efficiency of relatively wide-area
by flight control systems, which generally control coverage under the right conditions.
the camera remotely for optimum exposure,
mean that in some circumstances the image Rotary SUA are more useful for subjects where
settings chosen are inappropriate for accurate oblique imagery is required. Rotary SUA are able
results (Figure 70). In the example in Figure 70, to take oblique images as well as verticals. They
the shutter speeds were too slow. Small fixed- are more manoeuvrable than fixed-wing solutions
wing solutions generally provide a less stable and can carry larger sensors, but are slower and
camera platform, because continuous movement have a shorter battery life. They are typically
at speed is required to remain aloft. Shutter used for monitoring, inspection (Figure 71) and
speeds are therefore kept relatively fast to avoid visualisation, but may also be extremely useful if
blurring. In overcast conditions, this can lead to significant vertical elements (sides of buildings,
underexposed images. However, in most cases cliffs, etc) require recording, because these
the images process well enough, and the products can be poorly represented in vertical imagery.

< < Contents 59

 For ‘soft’ detail (eg the edges of ditches and
banks, which do not have unambiguous lines
defining them), GSD smaller than 40mm is rarely
useful in landscape archaeological contexts,
although for ‘hard’ detail (subjects that have clear
and unambiguous edges) it can be insufficient
for cleanly resolving these edges, and in that
case a GSD in the region of 10–20mm may be
more appropriate. In most cases a reasonable
compromise can be made given the size of the
smallest detail that needs to be resolved, the size
of the area to be covered, the computing power
Figure 71 available to produce a result from the inputs
Image taken from a rotary SUA for wall-top inspection. generated and the required output scale. If you
© Skyline Images
need to monitor change on a site over time, the
GSD has to be sufficient to allow detection of that
change, and the control network used should also
be measured with an accuracy commensurate
Rotary systems tend to be able to carry heavier with the size of change you wish to detect.
payloads, and thus generally better cameras or
other sensors. For imagery that is to be used for Rotary SUA are the platform of choice for most
site presentations, a more stable rotary system building survey work where aerial images are
is usually best and the acquisition of high- required. Wall-top monitoring, inspection and
definition (4K and higher) video is possible; this measured survey all require a stable platform able
can be processed photogrammetrically, as image to focus on the point of interest. Some SUA now
overlaps are very high, although they may be come with collision avoidance sensors, which
subject to either or both motion blur and rolling allows a constant stand-off from the subject to be
shutter problems. Rotary SUA can generally maintained. This feature can be particularly useful
operate with wind speeds up to 20km/h for high-level building survey.
(TSA 2015).
SUA are controlled and monitored from the
In general, SUA offer several advantages: ground. Most fixed-wing and some rotary
platforms offer the option of pre-flight planning
„ because of the lower altitudes at which of the swaths to be flown and the frequency
they are flown, they can avoid some weather with which images are taken, so enabling
conditions that make photography from a a good enough overlap to ensure even and
manned aircraft difficult photogrammetrically reliable results and
guaranteeing sufficient coverage of the site.
„ flying at lower altitudes also means that The software solutions available to carry this
they can provide higher resolution mapping out vary from manufacturer to manufacturer,
if required but the principles remain the same. It is a good
idea to enlarge the area of capture slightly when
„ they are a fast and flexible means of acquiring or procuring aerial imagery, in order to
acquiring data ensure that there is sufficient coverage of the site
right up to, and slightly beyond, the edges. In the
„ costs are relatively low for those recording UK, beyond visual line of sight (BVLOS) flying of
discrete sites on a regular basis and there SUA is not currently permitted (in some European
are many commercial operators competing countries this restriction does not apply), which
for business if you are procuring survey. means that, with a line of sight restricted to 500m

< < Contents 60

horizontally, the area that can be safely flown on 3.1.1.4 Other sensors
any single flight is restricted to 1km 2. In the last couple of years, lidar sensors that are
light enough to be carried by SUA have come onto
Some of the occasional imaging problems with the market. These sensors can capture higher
manned aircraft can be avoided by using an SUA. resolution lidar data than currently available
Haze, unless taking highly oblique shots, is not from a manned aerial platform, and retain all the
normally an issue because the flying altitude for advantages of being able to filter point clouds
SUA is capped in the UK at 120m above ground (for example to remove tree canopies and derive
level (AGL) and photography taken from a SUA a high-resolution DTM from the data). They also
is below cloud level (if it is not, the pilot should have the potential of integration with aerial
not be flying the aircraft because visibility is photogrammetric data, in a similar fashion to its
compromised). integration with terrestrial laser scanner data.

3.1.1.3 Kites and balloons Ground-penetrating radar (GPR) sensors that are
Kites and balloons have been used effectively light enough to be carried by a rotary SUA are
for aerial photography for a long time. Both also now available. These require a flight height
make efficient platforms for photogrammetric of approximately 1–1.5m AGL. They generally
work and have been used successfully for weigh about 1kg, and operate at a frequency
archaeological purposes (for examples see between 500MHz and 1GHz, with a manufacturer’s
Kite Aerial Photography). As this section is quoted ground penetration of approximately
concerned principally with image acquisition, 2.5m, although this will vary with frequency.
the special properties of kites and balloons are These sensors have not yet been evaluated by
less important; the aim is to obtain results similar Historic England, but obviously combining data
to those described for SUAs (particularly rotary derived from such sensors with that from other
platforms). A typical configuration of images taken sensors offers exciting possibilities for later
by a kite (in this case to examine a roof structure) analysis and interpretation.
is shown in Figure 72.

Figure 72
Image configuration taken from a kite.

< < Contents 61

 Desktop scanners scanners and have a random distribution.

Desktop scanners are cheap, easily available, The errors introduced by a desktop scanner
and make it possible to scan photographic will therefore influence interior orientation
prints when negatives are not available. estimates, thereby propagating errors
There are, however, some caveats to their throughout the photogrammetric workflow,
use in a photogrammetric workflow. They are adversely affecting the accuracy of the product
designed for non-photogrammetric users, and leading to high RMSE during, for example,
and are suitable for general use as far as aerial triangulation.
metric resolution (typically up to 1600dpi or
higher) and radiometric resolution (8 or more It is therefore advisable to scan negatives using
bpp) are concerned. The geometric precision a photogrammetric scanner; several makes
of desktop scanned imagery is, however, are available, albeit at a cost, and if accuracy
very poor (Calarco et al 2004; Mitrovic et al is a priority it might be worth outsourcing
2004) and brings with it the introduction of this element of the process. If scanning prints
largely unquantifiable and variable errors in with a desktop scanner cannot be avoided,
both axes in addition to the normal camera the probable effect on the accuracy of the
lens distortions already present in the product should be noted and made explicit in
prints. It is not therefore advisable to use the project metadata. Also be aware that the
a desktop scanner for scanning prints for print media, if not stored under good archival
photogrammetric purposes unless accuracy conditions, can become distorted, bringing
is not a high priority. They are, however, another unquantifiable error source into play.
extremely useful if outputs commensurate with
visualisation are all that is required. This holds When scanning prints, it is obvious that a
true for the scanning of any prints, historical very high scanning resolution cannot
or otherwise. produce a higher resolution product than
the original image but will produce images
The precision of different makes and types
of non-photogrammetric scanners is very
variable, and they exhibit both systematic and
random errors, typically up to 1.1 pixels but
with spikes of up to 3.5 pixels (Calarco et al
2004; Mitrovic et al 2004). Initially systematic
errors may not seem to present too much
of a problem, as theoretically they can be
compensated for to some extent by calibration,
where this is possible. This can require the
use of a precision-etched optical glass plate
to quantify the errors, but experiments have
shown that even after calibration some
systematic errors remain, demonstrating that
the error pattern is not a simple linear one and
is inconsistent between captures (Calarco et al
2004; Mitrovic et al 2004).

Photogrammetric scanners produce scans Figure 73

with error sigma values that are approximately Aerial image with fiducial marks and
10 per cent of those exhibited by desktop other metadata.

< < Contents 62

 that take longer to process. There is little A single mask can be used across all the
to be gained by scanning prints with a images provided the inputs are consistent in
photogrammetric scanner. their offsets; this can be a problem with images
scanned from prints, which will therefore have
If you are procuring scans of prints from an to be masked individually. Some archives do
archival source, remember to specify that not allow the use of scanners but do allow
the removal of fiducial marks or any other photographic prints to be re-photographed;
information from around the edge of an this is an option but will of course bring
image is prohibited (Figure 73). These must in its own distortions. Prints should be re­
not be cropped but masked on the inputs, as photographed square-on using the longest
the whole of the input image is required to focal length lens possible.
estimate the lens distortion characteristics.

3.1.1.5 SUA regulatory framework annual hours flown and the environments in
All commercial SUA operators in the UK must hold which the SUA are typically operated. In addition,
a valid CAA Permission for Aerial Work licence. In permission must normally be sought from
order to get this, SUA operators must undertake the landowners. Permission must be sought
one of a number of approved training courses, before low-level flying over all English Heritage
gaining either a Basic National Unmanned properties.
Aircraft Systems Certificate – Small (BNUC-S)
or a Remote Pilot Qualification – Small (RPQ-S) SUA regulations in the UK are in the process of
for their specific SUA type. Once this has been change: for the latest information, visit the CAA
achieved, a company must additionally submit website. Historic England does not and will not
an SUA operations manual for each of its aircraft use contractors that are unable to demonstrate
(which must be re-submitted annually) and an that they hold the appropriate CAA permissions
SUA technical manual to the CAA, along with an and public liability insurances, and who have not
annual fee. The CAA categorises SUA by weight, undertaken a full risk assessment and method
with different classes having different restrictions statement for a planned flight.
on when and where they can be flown; separate
permissions are required for flying heavier SUA, 3.1.2 Analysis
and they require more experienced operators. The most often used product in archaeological
landscape analysis is a DEM, usually in
Non-commercial operators (those who undertake conjunction with an ortho-image. There are many
work without accepting ‘valuable consideration’ ways of pre-processing the point cloud data
for it) still need to be aware that safe and to provide a DEM for archaeological landscape
responsible operation is paramount and a legal analysis. Some photogrammetric software allows
requirement. An SUA can be heavy, moves at the data to be filtered to remove buildings and
speed and has the potential to cause trees. This is often not based on characteristics
considerable harm or damage if control is encoded in the data, as may be the case when
compromised in any way. dealing with full-waveform lidar data, but carried
out by dividing the dense point cloud into a series
All operators should ensure that they have of ‘cells’, making a local search of those cells to
adequate insurance before they undertake a establish a likely ‘ground’ level and filtering out
flight. Insurance premiums vary, and the growth of data in any particular cell that does not conform.
the technology has meant that more insurers are Vertical intervals between the ground surface and
moving into the SUA market. Insurance premiums other data can be specified to enable the removal
are in general affected by operator experience, of, for example, tree canopies or roofs. The more

< < Contents 63

laborious process of manual classification is an commercial photogrammetric software
option, in which case points are selected by hand packages permit the export of a DEM as well as
and assigned a class. This is often the easiest the TIN, and a DEM can be produced in an
way to identify water bodies, for example. More open-source workflow by passing the dense
sophisticated products, such as Erdas Imagine, point cloud or mesh to a separate piece of
are able to undertake more complex classification software such as Meshlab or Cloud Compare, for
of photogrammetric point cloud data based on a scaling, orientation and export in an appropriate
variety of characteristics, including colour. format for further analysis if required. Both of
these examples of software have a range of
In the case of medium to large areas, commonly powerful analytical filters that can be used if a
the subject of landscape investigation, the GIS is not available.
products are most effectively dealt with in a GIS,
which will typically involve exporting the 3D data Perhaps the most accessible and obvious ways
from photogrammetric software as a raster image. of using GIS to extract useful archaeological
It is often economical to use a specialist viewing information from the DEM are hillshades, slope
package such as QT modeller, or a visualisation analysis, contouring and, when there is more than
toolbox such as RVT, as an intermediate step. one data set available for comparison, principal
While these softwares are aimed at those using component analysis (PCA), although many other
lidar data, visualisations can often be done ways of dealing with such data are available. At
more quickly outside a GIS and then imported Historic England, the analyses mentioned have
as 2D raster images for further processing. GIS is proven to be the most fruitful outputs for field
commonly used not only because of the analysis teams to use, largely because they are relatively
that can be performed on the DEM, but because it straightforward to achieve for non-GIS specialists
allows integration of the photogrammetric output and show the data in a readable and interpretable
with spatial data derived from a large number of fashion. These analyses can be undertaken in
other sources, and examination of the interactions most GIS, both commercial (for example ESRI’s
between these data. ArcGIS) and open-source [for example QGIS,
GRASS, the Lidar Visualisation toolbox available
When exporting to a GIS, remember that the from the Arcland website and the popular Relief
raster image (DEM) is a regular gridded format in Visualisation Toolbox (RVT). The resolution of
which each pixel is usually square and represents the outputs chosen is dependent upon several
a regularisation of the more randomly distributed factors, including the resolution of the inputs
dense point cloud or the nodes of the mesh from (that is, the GSD), the size of the smallest details
which it is derived (usually, in its raw form, a TIN). that require representation, and maintaining
While this is not normally problematic, it does usability of the data: a large area covered at high
involve a small amount of smoothing in a similar resolution can generate enormous files quite
way to that involved when processing lidar data quickly, and hardware limitations may mean
into raster images (The Light Fantastic: Using that the outputs have to be broken down into a
Airborne lidar in Archaeological Survey). While number of smaller, seamless raster tiles to allow
a TIN may be imported into a GIS, and it has the reasonable processing times, especially on less
advantage that the data is not interpolated to the powerful computers.
same degree, TINs are generally more difficult
to process and typically require conversion into While it is sometimes useful to increase the
a raster surface before second-order derivatives resolution of DEM beyond its native GSD, it
such as hillshades can be extracted or other is fruitless to increase the resolution of an
forms of analysis undertaken. A useful summary ortho-image beyond this: you would simply be
of the advantages and disadvantages of dealing increasing the number of pixels in the output
with point cloud and raster surface data is given without a corresponding gain in resolution.
in (The Light Fantastic: Using Airborne lidar in If you do increase the resolution of a DEM for
Archaeological Survey, 11). Most off-the-shelf output, bear in mind that intermediate points

< < Contents 64

will be interpolated from surrounding values and „ Office: preparation, research.
that, without the use of breaklines, this can also
introduce smoothing into the raster grid. Whether „ Field: reconnaissance, including full
this is acceptable or not will vary in different field walkover and formulation of
situations and with the type of detail or variation initial interpretation.
you are attempting to resolve.
„ Field: image acquisition.
At higher resolutions you should also consider
that variations in vegetation can produce noisy „ Office: image processing, DEM and
data: with a GSD of circa 40mm, variations in the ortho-image production.
length of grass over a site can have a significant
and possibly detrimental effect on the DEM. „ Office: GIS-based analysis and generation of
This sort of noise is reduced considerably with interpretation aids such as hillshaded DEMs.
increased flight height, and is almost never an
issue with photography acquired from a manned „ Office: vectorisation (and hence
aircraft because such very small details are interpretation) from these outputs, usually
rarely clearly defined. As mentioned in section as top and bottom edges of slopes or
2.1.9 Multispectral imagery, these variations feature outlines.
can sometimes be useful, especially with
regard to crop heights, for identifying potential „ Field: checking of analytical outputs,
archaeological features, although their causes can refinement of linework, refinement of
be manifold and care should be exercised in their interpretation.
interpretation. Differences in grazing regime, for
example, may mean that one part of a field has „ Office: production of finished interpretative
shorter grass than another, and particular care illustrations based on revised outputs.
should be taken when comparing DEMs of the
same place from imagery shot at different times This workflow is explained with reference to a real
of year. It is also worth noting that vegetation site in case study 2.
heights can have a significant effect if you are
integrating or comparing a photogrammetrically The importance of reconnaissance cannot be
derived DEM with a lidar-derived one: the overstated, and serves several purposes. For the
photogrammetric model will often appear archaeologist undertaking the interpretation, it
raised above the height of the lidar model, this is an opportunity to become familiar with the
difference being a function of vegetation height, site and to start the process of understanding
which is usually filtered from lidar data (Green et the features that are visible, the stratigraphic
al 2014). and temporal relationships between them and
their spatial and cultural context. It is also an
Out of the box tools are available in most GIS opportunity to assess potential safety issues that
for performing slope, hillshade, contour and should be communicated to the subcontractor
PCA analyses of DEM data. Other, more complex (if one is used), to consider where GCPs may be
methods of analysis are possible, but for rapid most usefully placed, to consider the extent of the
interpretation those listed here provide a sound area that will require coverage, and to decide on
basis. Usually, a combination of products, the necessary GSD to resolve the required detail
strengthened with a high-resolution ortho-image, successfully at the desired scale of output.
provide the best starting point for extracting
useful interpretative archaeological plans. For Large, repetitive linear features, for example ridge
landscape work, the workflow that has proven to and furrow, land drains and stratigraphically
be most effective is as follows. straightforward or discrete features, can be
mapped directly from the DEM rather than in the

< < Contents 65

field, although a degree of field verification is For aerial reconnaissance work, it is often
necessary. This method can save a great deal of sufficient to produce an ortho-image alone
time for field surveyors who would otherwise have rather than a DEM, simply to locate the feature(s)
to map these with GNSS; they can then devote correctly and allow its digitisation
more time to interpreting more ambiguous or and interpretation.
complex areas. Aerial imagery can also be used
to record detail in areas that are not otherwise
accessible (for example on military ranges), 3.2 Terrestrial photogrammetry for
although care should always be taken to secure buildings and structures
and observe access rights and respect rights to
privacy, etc. The SfM–MVS workflow provides most of the
advantages of traditional photogrammetric
The process of digitisation from a DEM and slope products for the survey of buildings and structures
analysis in CAD is shown in Figure 74. Firstly, with the addition of new benefits. SfM outputs are
field drains and water courses are added. Next, sometimes used on their own but, like the data
top and bottom edges are digitised, followed by surveyed by other methods, are usually a means
hachures to make the slope direction clear. In the to an end and need conversion into a useful
final image the underlying hillshade of the DEM product. The most commonly used products in
has been removed for clarity. The linework is now the recording and presentation of buildings and
ready for field revision followed by the production structures are ortho-images and 3D models. DEMs
of the final interpretative drawing. and products useful in other areas are less often

Figure 74
Digitising archaeological interpretation from DEM
and slope analysis.

< < Contents 66

employed unless assessing a relatively localised One important difference between the data
deformation; more general deformations can be derived from tracing linework from an ortho­
assessed via analysis of the full 3D model. image and a traditional photogrammetric output
is that the ortho-image-derived linework will
3.2.1 Ortho-imagery be 2D only. For true 3D linework, detail must
A scaled ortho-image derived from be traced directly from the 3D model, and few,
photogrammetry can be used as the basis for: if any, SfM–MVS-based packages allow this at
present. In most cases, however, this does not
„ tracing 2D linework in CAD present a problem because the most commonly
disseminated outputs from archaeological
„ as a product in itself (providing a metric building survey still focus on a ‘flat’ 2D product,
record of materials, condition, colour presented as hardcopy or in a .pdf or CAD format.
and arrangement, including defects such
as cracks) There are alternative methods for producing
3D linework as output (which was a standard
„ a combination of the two. product with stereo photogrammetry), and these
usually involve importing the dense point cloud
The point densities required from SfM-generated into software more commonly used for 3D laser
data for ortho-photographic processing, at least scanner data (or indeed many CAD packages) and
that procured by Historic England, are given digitising the detail using those means, although
in Andrews et al (2015). Because ortho-image some software dedicated to this task is available.
production is contingent upon having a DEM, the All of the problems usually associated with
accuracy of which will determine the accuracy of digitising from point clouds remain, and it can be
the ortho-image, the spacing of points required in a lengthy and difficult process to get the process
the DEM to be used for generating an ortho-image, right with complex geometry, although there
and the maximum permissible GSD at the same will be a large number of high-quality reference
scales, is given in Table 4. images to resolve points of doubt.

Recording the exterior of a building

Output Maximum Maximum GSD photogrammetrically is faster in the field than
scale point spacing using the more ‘traditional’ combination of a
1:10 1mm 1mm TST and hand measurement, but this is to some
extent compromised by the extra time required
1:20 5mm 2mm
in the office to produce the ortho-image (and
1:50 10mm 3mm digitise from it if required), as well as less time
1:100 10mm 10mm spent in the field looking at, and understanding,
the subject. The amount of time saved generally
1:200 20mm 20mm
increases with scale; you will save more time
1:500 40mm 40mm overall if you are required to produce drawings
at 1:20 than at 1:200, because the frequency with
Table 4 which measurements must be taken with the TST
would be greater, and it is far easier to digitise
from the ortho-image in CAD at the required
intervals than in the field). For accurate work a
In practice, given the high resolutions that can be TST can be used to add control measurements,
obtained with most modern cameras, these GSDs usually to easily identifiable points of detail on
will usually be exceeded. the elevations (for example corners of windows
and ventilation grilles). The TST is invaluable if
you plan to integrate more than one elevation in

< < Contents 67

the same coordinate system, or if the elevation 3.2.2 Elevations
has significant projecting detail (for example Recording a single building elevation is perhaps
porches and colonnades). Control points and their the most straightforward use of photogrammetry.
distribution are discussed in section 2.3 Control. In general, the best approach is to attempt in part
to mirror the aerial application of the process, by
Detail that cannot be seen in the input providing a series of overlapping images covering
photography cannot be accurately reconstructed the entire elevation from left to right, bottom to
in the photogrammetric model. Thus it is top, but braced with additional oblique images. If
important to ensure that the tops of sills, bay nothing else, a methodical approach encourages
windows, etc, are recorded in some fashion, even completeness of coverage (see section 2.2 Image
if by hand measurement on a few examples, which arrangement). This is most easily achieved, in the
can then be incorporated into the final drawing case of smaller buildings (less than three storeys),
if line output is required. If the ortho-image is by using a combination of photographic tripod
to be the product, a way of filling any gaps must (for the ground floor) and a mast at different
be found (by taking imagery that covers them) heights (for the upper floors). Care should be
or their presence indicated (Andrews et al 2015). taken to ensure that there is enough side lap
As with all methods of drawing production, one between the runs to allow their relative positions
single survey technique will rarely provide all the to be computed successfully. In many urban
information required for a finished product. situations, the narrowness of streets means that
stand-off is compromised. In such situations
the use of a wider angle lens (for example 18 or
24mm) can be helpful, as fewer shots need to be
taken. Figure 75 shows an example of the field
of view of 24mm and 50mm lenses from same
position. Roofs, however, can still present a
problem. Common workarounds include:

„ increasing the stand-off and using a zoom

lens to fill in any roof detail not visible from
the ground

„ if the stand-off cannot be increased,

attempting to see as much of the roof as
possible by moving to the sides and taking
fill-in photography from there

„ arranging access to a building opposite and

taking the necessary photographs from its
upper floors or roof.

In some cases it may only be possible to

record the roof properly using a TST or hand
measurement, either by gaining access to the
roof itself and surveying it separately (but in
the same coordinate frame as that used for the
photogrammetric recording of the main elevation)
Figure 75 or by surveying significant features (rooflines,
Images of the same elevation taken from the same chimneys, plants, etc) from the ground if they
standpoint using lenses of different focal lengths.
are visible. This may also be necessary in the

< < Contents 68

case of very tall buildings fronting onto narrow
streets. Highly oblique measurements with a TST,
although undesirable, may be necessary as they
are likely to be more accurate than those derived
from photographs. Points on roofs taken with a
TST can also be used as photogrammetric control
if sufficient image coverage has been obtained.
In a more rural or open setting, infill photography
of roofs taken from a SUA or kite can provide the
best means of obtaining the necessary data to
complete the photogrammetric model without
additional use of the TST. In urban environments,
special permissions are required to fly SUA and
road closures may be deemed necessary.

Where more than one exterior elevation is to be Figure 76

recorded, the best results are achieved if a single Poor representation of ironwork in an ortho-image
from general shots of an elevation.
control network is established around the building
(using a TST), thus allowing all of the elevations to
be recorded in a common coordinate frame. This
also makes the process of drawing production in
CAD considerably easier.

Interiors are more easily dealt with, and very

good results can be obtained in most
circumstances provided there is sufficient light
(see section 3.4.2 Lighting). When photographing
wall paintings, for example, use of colour
reference cards is essential. Do not forget to
photograph the floor as well, if possible, or the
model will remain incomplete if it is required for
visualisation purposes.

3.2.3 Problem areas

Certain sorts of detail (notably decorative
ironwork) will usually reproduce poorly in a
SfM-derived model and ortho-image. The general
positions of such objects will be visible but the
overall model will usually be incomplete (Figure
76). Fill-in photography focusing on the problem
area is one potential solution but, if a line drawing
is the desired product, it is usually best to record
such things using hand-measured sketches, draw
them up in CAD and add them to the drawing in
the appropriate position using the ortho-image
as a guide. Complex curved metalwork structures, Figure 77
especially those that are highly specular or Extraneous geometry produced when photographing
windows.
coated with glossy paint, will also model poorly;
for such cases you will need to use a different
method of recording.

< < Contents 69

Glazed windows can also be problematic, and In a similar vein, if possible it is usually worth
usually produce large numbers of extraneous masking out the sky from the input images before
polygons in the final model, resulting from both processing. Not only does the sky have no useful
reflections and objects visible through the window points for matching images together, it can also
(Figure 77). If possible, it is often better to mask generate large quantities of extraneous geometry
windows out of the input photographs before that require subsequent removal in a similar
processing; the necessary detail can be added to fashion to glazing-related artefacts (Figure 78).
the final drawing later by other means. If masking
is not an option, extraneous points should be 3.2.4 Integration with 3D laser scanning
removed manually from the dense point cloud Point cloud data derived from photogrammetry
before generating a polygon mesh, if that is can be integrated successfully with data generated
required. If you do not manually edit the dense by a 3D laser scanner ('hybrid modelling') if that
point cloud, but set an upper polygon limit for the satisfies the metric requirements of the project
model when you generate a mesh, a considerable and is on the same control system. It is often
number of the polygons generated are likely to possible to obtain photogrammetric data from
be ‘junk’ data associated with reflections, etc; areas where it is not possible to get a line of sight
these will have to be deleted, which will mean from a scanner, for example by using a camera on
that you have fewer polygons from your ‘budget’ a photographic mast. There is a lot of software
representing detail that is of interest. available for the processing of point cloud data
derived from scanners and it is beyond the scope
of this guidance to detail them all. One issue to
note, however, is file formats. Many packages
designed to deal with terrestrial scanners can
now import the ‘universal’ E57 format; prior
to this, nearly all scanner manufacturers only
used proprietary formats, which meant that
merging data could be problematic. Some
photogrammetric packages do not currently
permit the export of point cloud data in the E57
format, but instead use formats useful for other,
largely visualisation focused, purposes (for
example Wavefront .obj and Stanford .ply). They
do usually permit export in a lidar .las (or .laz)
format, as many of them are focused on aerial
mapping, which can be useful if you are planning
to integrate your aerially derived point clouds
with lidar data.

As described elsewhere, data derived from

the scanner can be used as a control for
the photogrammetric data. Some software
packages, for example RealityCapture, are now
able to integrate scanner data directly with
photogrammetric data. This uses the data from
both inputs and combines them into a fully
integrated product derived from both sources.
If you are not able to use such a package, it is
Figure 78 possible, in Agisoft Photoscan Pro, for example,
Masking sky in input images to reduce extraneous to process the laser scanner data in a separate
geometry at the processing stage.
software package (for example FARO Scene or

< < Contents 70

Leica Cyclone), export a model, import it to the
photogrammetric package, and texture it using
the (previously aligned) high-quality images
from a better camera than that provided on the
scanner. The data from both sources needs to
be registered accurately in the same coordinate
frame. This processing workflow can be useful
because most terrestrial laser scanners are poorly
equipped with regard to cameras and textural
information is therefore often lacking.

3.2.4 Tripods
A photographic tripod is an essential piece
of equipment for terrestrial photogrammetric
imaging. By stabilising the camera, smaller
apertures (and hence greater depth of field) are
possible because they can be compensated for
by using longer exposure times, if external lighting
rigs are not available, for optimal exposure. A
tripod should be used as the default, and hand­
held use of the camera resorted to only when
necessary. If shutter speeds slower than about
1/60second are used, a tripod is required to avoid
blur. The extra time taken moving the tripod
around is more than compensated for by the
increased chances of obtaining sharp images,
which will always process better and give a
superior product.

3.2.5 Mast or extendable tripod

As normal photographic tripods are only helpful
up to the height of the photographer using them,
other solutions may be necessary in terrestrial Figure 79
photogrammetry to raise the camera higher, for Use of an extendable photographic mast.
example by using masts or extendable tripods
(Figure 79). The first rule with using a mast for
photography is to do so safely. Keep away from
power lines or other obstructions, and do not use be undertaken by twisting or moving the mast.
it if you are unsure about your ability to control A simple tilt head greatly reduces the weight
it in windy conditions. Keep well away from at the top of the mast and is cheaper. Using
windows if possible. It is usually best to have two lighter cameras is also desirable, although not
people available for hand-held mast photography, always possible.
thus reducing the chances of a slip that can have
unpleasant consequences for the subject of your Use of a mast (typically constructed from glass
photography, yourselves and/or your equipment. fibre, aluminium or carbon fibre) means that there
is movement of the camera at the top of the mast,
If you are using a mast, you do not require especially under windy conditions. If possible,
an ordinary pan and tilt photographic head therefore, try to avoid shooting in very windy
(commonly used on tripods), as panning can conditions, and even under calm conditions

< < Contents 71

keep the mast as still as possible and allow the at ground level (for example dwarf walls and
vibrations caused by moving it from one location archaeological excavations) the photographs
to the next to settle before taking photographs. they provide can easily be used to generate an
Faster shutter speeds will be required to avoid ortho-image, which is difficult to derive from more
blurring images using this method, because of the oblique photography taken from ground level. In
inherent instability; masts are usually hand-held addition, the area covered by each image is larger,
at the base, but even if mounted on a vehicle or so in most cases fewer images are required. The
sturdy ground anchor there will still be movement ortho-images can be used partly for the derivation
at the top except on the calmest of days. It may of vector plans, by taking the ortho-image into
be necessary to run the camera in shutter priority CAD or GIS software and digitising detail from
mode, and select a shutter speed of 1/60 second it. Although true vertical shots are impossible
or faster if conditions allow. (the mast itself gets in the way), near-verticals
can be taken, and the use of L-plate adapters,
As well as providing a more stable base, vehicle- which allow the camera to be orientated in either
mounted masts can be much taller than hand­ portrait or landscape mode, can be helpful. In
held ones, and are typically raised and lowered building survey, at least of smaller structures,
hydraulically or using an electric motor, which masts can be used to provide shots that fill in
considerably reduces the amount of physical work detail that is not visible from ground level.
needed to use them. Faster shutter speeds can
be compensated for by using wider apertures, A remote trigger for the camera is essential
so faster lenses (those that will open to wider when using a mast, and a number of different
apertures, thus letting in more light) are often options are available that vary considerably
desirable, especially in low-light conditions. You in functionality and price. The Camranger, for
need to be aware of the corresponding reduction example, gives you full control of the camera
in the depth of field, but this is not usually an settings while it is aloft, uses the camera’s live-
issue given the stand-off from the subject that view to show what the camera is pointing at, and
can be achieved. Hand-held masts usually cover can be controlled using a mobile phone or tablet
a range between 5 and 12m, after which they can at ground level via an app (this works for both
become unmanageable, while vehicle-mounted Android and iOS as well as on a laptop); it works
masts can extend the range to 40m. with most popular DSLRs. Other similar solutions
are available. Many camera manufacturers offer
Masts are extremely useful for a number of their own apps for products that offer similar
applications. For subjects that are more or less capabilities (Sony’s PlayMemories Mobile, for
example, can be used with their popular Alpha
7 range of cameras, and offers control of some
functions of the camera via a mobile app).
Communication between camera and ground
unit is typically via a wireless network connection
between the two. There are many other solutions
that offer a wireless remote shutter capability,
which can be useful under some circumstances,
although being able to preview an image is not
always possible and control of camera settings
can be problematic. Having to change camera
settings manually can take some time when using
a mast, as the camera has to be brought back
down to ground level for the adjustments, and
this can be particularly frustrating on days with
Figure 80 changeable weather conditions. When using a
Rain sleeve on a camera for use in wet conditions.

< < Contents 72

mast on wet days, a rain sleeve or cover for the and relationships between features, can be
camera (Figure 80) is a worthwhile investment easily incorporated into a GIS or CAD system for
(many are available), as is the use of a lens hood generating plans, and can be digitised to derive
to minimise water droplets getting onto the lens. vector outputs. Sections can be recorded in this
way. Similarly, DEMs can be used, for example
to map deformation in mosaic floors (Green et
3.3 Terrestrial photogrammetry al 2014). 3D textured mesh models of particular
for excavations features or areas of a site can be very useful: a
model of a site can be sliced to provide profiles
Provided the principles outlined in other in any orientation. In all cases, comprehensive
sections are adhered to, the potential coverage is key and a methodical approach is
applications of photogrammetry for advised. Use of an accurately measured common
archaeological excavations are many. SfM–MVS control system will allow you to combine the
approaches offer a more affordable method for results from adjacent or overlapping trenches
3D recording than terrestrial laser scanners, excavated at different times into a single model,
record colour and textural information better, are allowing the spatial relationships of objects
faster to acquire and provide outputs of a suitable and areas not seen at the same time to be
accuracy (Doneus et al 2011; McCarthy 2014; appreciated in three dimensions. Taking repeated
see case study 6). In common with other mass- photogrammetric surveys as an excavation
capture data acquisition methods, however, the progresses enables the composite reconstruction
outputs are metric but still require interpretation, of layers, allowing, for example, virtual sections to
unlike, for example, a hand-drawn section, which be generated.
transmits the interpretation through the process
of selection at the point of capture. You should include a colour index card and/
or grey card in the images, and you will need to
Archaeological excavations are generally visually re-shoot if lighting conditions change during the
texture-rich, and matching images is not usually photography, to enable a reasonably consistent
problematic. Wet surfaces or puddles on a output. Control is usually measured with a TST,
sunny day are likely to reflect light and appear and will therefore be accurate to within ±3mm
shiny, so under these circumstances matching (reflectorless) at ranges typical for archaeological
can be compromised or noisier data produced. sites. As for other applications, the control can
Puddles should, as in the case of ‘traditional’ site be used to both refine the orientations of the
photography, be removed before imaging cameras and to provide absolute orientation for
if possible. the model with respect to the site coordinate
system. If using SfM for an archaeological site
Perhaps most usefully, ortho-images offer an or feature without a TST to provide control, it is
excellent means of recording plan information, essential you include scale bars in the images.
especially when footfall across an area should A small amount of forethought means you can
be minimised. The delicate and complex wooden produce models that are not only visually
platforms at Star Carr, North Yorkshire have been pleasing but also metrically accurate.
recorded using this method over the last few years
(case study 5) as well as, for example, recent
excavations at Must Farm, Cambridgeshire. The 3.4 Terrestrial photogrammetry
GSD achievable (even using a photographic mast) for smaller objects
with a reasonable digital camera is more than
adequate for recording at a scale of 1:20 The principles that must be applied when imaging
or smaller. smaller objects are the same as for larger objects,
but some of the equipment and techniques that
As with building survey, an ortho-image provides can be used will vary.
an excellent representation of colour, texture

< < Contents 73

3.4.1 Macro lenses A shallow depth of field is much less of an issue
True macro lenses provide a magnification factor when shooting a largely planar subject: purpose-
of at least 1 (or a scale of 1:1) at their closest built macro lenses are ‘flat-field’ lenses, so the
focus setting. This means that, at a scale of 1:1, centre and edges of a subject that is planar will
the object will appear on your camera’s sensor at all be in focus. Most small objects that you want
the same size as it is in real life. Magnification is to model, however, will not be planar. Several
determined by the focal length of the lens and workarounds are available to reduce the impact
the focusing distance: the closer you can focus, of a shallow depth of field, but not all of those
the more magnification a lens of given focal useful to, for example, a wildlife or product
length will be able to achieve. Macro lenses are photographer (such as focus stacking or other
capable of much closer focusing distances than image manipulation techniques) are suitable
ordinary lenses to achieve this magnification; if for photogrammetric purposes. All workarounds
you are using an ordinary lens, you will be well involve some compromise.
inside the minimum focusing distance for your
lens before you can achieve a reproduction ratio Decreasing the aperture: ‘stopping down’, or
even close to 1:1. reducing, the aperture (to, for example, f/16 or
f/22) has the net effect of increasing the depth
Different focal lengths of macro lens are useful of field but will also, after a certain point, cause
under different circumstances: generally, longer a loss of sharpness as a result of diffraction (the
focal length lenses allow a greater stand-off ‘diffraction limit’). Stopping down also requires
from the subject; thus, for example, a 40mm much slower shutter speeds to compensate for
or 60mm lens allows you to get very close to the lack of light reaching the sensor, higher ISO
a subject, while a 200mm lens allows you to values, or the use of external lighting.
have a reasonable distance between the lens
and the subject with the same result. With Arranging the subject so that it is as parallel to
photogrammetric use of macro imagery, the main the sensor as possible: subjects that are long
practical difference is that a longer focal length and narrow will not image well if shot from either
lens allows more lighting to be placed between end. Most of the subject, given the shallow depth
the camera and the subject, which can be useful of field, will be out of focus (Figure 81). It makes
for the reasons outlined. sense, therefore, to arrange the subject as close
to parallel to the plane of the camera sensor
One characteristic of macro photography, as possible in each shot, allowing most of the
and the main issue when using it subject to be in focus in each image. In practical
photogrammetrically, is that the depth of field terms, this involves rotating the subject about its
tends to be very shallow at ‘normal’ apertures. long axis while it is being photographed.
Thus, for example, when using a 100mm macro
lens at f/2.8 on a camera with an APS-C sized Increasing stand-off: this has the effect of
sensor, the depth of field is approximately 0.6mm: lessening the magnification at the expense of
only those parts of the subject that are 0.3mm in having the subject fill less of the frame. This works
front of or behind the focus point will be sharp for applications of macro photography where an
(in macro photography the depth of field remains image can be cropped for presentation purposes;
symmetrical about the focusing distance). for photogrammetric imaging it has the effect
Decreasing the aperture to f/11 extends the of reducing the resolution achieved but it can
depth of field to approximately 2.6mm. Smaller provide additional background areas that can be
apertures will increase the depth of field further used for image matching. The required resolution
(but at a cost). It is therefore best to use a tripod, of the product will have a bearing on whether this
focus macro lenses manually and use the live workaround is acceptable or not. External lighting
view, if available, to refine the focus point. can be used with increased stand-off.

< < Contents 74

Using a different camera: cameras with smaller
sensors can achieve better results with regard
to frame filling, depth of field and effective
detail when imaging small objects, albeit at
the expense of all the factors mentioned in
section 2.1.2 Resolution and sensor size. In
common usage, the term ‘macro’ is often used to
describe close-up photography, rather than ‘true
macro’ photography. In close-up photography,
magnifications of 1:10 or so can be achieved
and the subject can still fill the frame. The size
of object that can fill the frame is dependent on
the sensor size of the camera used: the smaller
the sensor size, the smaller the object. Thus an
object imaged at 0.5× magnification will occupy
less of a full-frame 35mm sensor than the much
small sensor of a compact camera, so, effectively,
a smaller sensor allows smaller objects to fill the
frame. The magnification factor achieved is often Figure 81
expressed as a 35mm equivalent magnification Shallow depth of field causing problems with the ends
of ‘long’ objects in macro photography.
for smaller format systems. The depth of field is
increased when using smaller sensors at the same
magnification; the smaller sensors on compact
cameras offer an increased depth of field over
full-frame 35mm cameras, and this can be a Setups more commonly associated with product
significant advantage, although tempered to some photography can be used, such as light tents. The
extent by the slightly reduced image quality in object is placed inside the tent for imaging and
other areas. lights are used outside the tent, which scatters
the light, allowing diffuse and even coverage if a
3.4.2 Lighting number of lights are used (Figure 82).
Given the small apertures that often need to be
used, lighting is important when imaging very 3.4.3 Turntables
small objects. Even lighting is as essential when Turntables can be used with small objects to
imaging smaller objects as larger ones. For very allow the camera to remain relatively static while
small objects, a ring flash unit can be extremely the subject is revolved between each exposure,
useful and is not necessarily expensive. LEDs can generating the effect of circling the subject with
provide a very cold light, and the white balance the camera. Specialised photographic turntables
should be adjusted accordingly. A colour index are available, but in practice turntables designed
card should always be included, in one shot if the as a cake stand (a ‘Lazy Susan’) can work just as
lighting conditions remain consistent and in more well provided they are stable and the object not
than one if the lighting conditions change. If a ring too heavy. This method can shorten the image
flash unit is not available, a flash unit attached capture phase considerably because the object
to the hot shoe of the camera can be used; a is moved rather than the camera. It is advisable
flash diffuser should then be employed, and it is to shoot more than one ‘ring’ of shots round the
often useful to point the flash unit away from the subject, for example by increasing the camera
subject (for example upwards) and bounce the height for each ring (Figure 83 shows an example
light from a reflective surface to provide more of an image arrangement; see section 2.2 Image
even lighting. If possible, the best option is to use arrangement). Backgrounds should be kept as
external or studio lighting. neutral as possible.

< < Contents 75

 Control for small objects is discussed in section
2.3. If you are using a fine gridded film placed
under the subject, it should be clearly visible in
all images; if you are using scale bars, make sure
you include them in sufficient images. Given the
very high GSD achievable in macro or close-up
imagery, bear in mind that on some scale bars
the width of the graduations on the scale itself
can be many pixels wide (Figure 84), placing
a constraint on the accuracy with which the
object can be scaled. Callipers often have much
finer graduations than scale bars, and make an
excellent alternative if available. Alternatively,
as discussed in section 2.3.2 Sources of control
data, coded targets or scale bars can be used.

When modelling an object in full 3D, it is often

necessary to capture the upper and lower surfaces
separately, with sufficient overlap between
the two, and merge the parts later to form a
single model, as discussed in section 2.2 Image
arrangement. In some cases, an additional set
of imagery covering the overlap, and common
to both sections of the model, can ameliorate
the effect of noisy data at the edges of each half.
This noise is typically the result of a loss of focus
(because of a shallow depth of field) towards
Figure 82 (top) the edges of the models, creating a ‘seam-line’
Use of a product photography tent for lighting small where they join (Figure 85) if this is not rectified
objects evenly.
by additional imagery. Use of mirrors to image
Figure 83 (above) inaccessible areas of subjects (Mallinson and
Use of a turntable for imaging small objects. Wings 2014) is not advised because the geometric
properties of the image are compromised.
Alternatively, defects at the joins can be removed
using external mesh editing software, but with
a consequent loss of reconstruction accuracy in
that area of the model. It is often necessary to
use distinctive points on the objects themselves
to match between different models if there is
not sufficient overlap between the two for an
automatic tie point-based alignment from the
input imagery.

Making models of small objects

photogrammetrically can be time consuming,
especially at the image capture phase. While
this is usually not problematic with one or two
objects, it is worth considering whether an
Figure 84 alternative approach would be more economical
Scale bars photographed with a high resolution camera. if large numbers of small objects are to be

< < Contents 76

modelled. High-resolution 3D laser scanning,
for example, is not affected by depth of field
issues, and for collections of objects such
as bones, arranging access to a computed
tomography (CT) scanner, if possible, can
be a more fruitful approach, albeit offset by
considerations of cost and a lack of colour texture.

3.5 Textured 3D models

Objects are the most common subject that

requires a 3D textured model. If the desired Figure 85
output is intended primarily for visualisation ‘Seam’ lines appearing in a model where there is
insufficient overlaps between halves.
purposes, the metric requirements of the
model can be lower than those required for
other purposes. A well-designed workflow for
dealing with the data should ensure accuracy
throughout the process, but the polygon counts There are many other ways of disseminating
in the final models can probably be reduced 3D model data. For email distribution, 3D PDFs
drastically in order to reduce file size, and so are popular and allow other users without access
aid transmissibility and accessibility. Reduced to sophisticated 3D software to view, label,
polygon counts can often be ‘disguised’ by the section and measure low-resolution 3D models
effective use of good-quality textures. (Figure 86).

Figure 86
Annotation and measurement capabilities in a 3D PDF.

< < Contents 77

Online resources such as Sketchfab offer a 3.6 3D printing
useful service for sharing and viewing models
both publicly and privately, and 3DHop provides Once models of objects have been produced,
another very useful open-source method of there is the potential to produce 3D-printed
sharing even high resolution models on standard replicas or scaled representations, for example
web pages. Other sites offer more comprehensive for educational or display use. Such models
services, for example Drone Deploy, which both need to be ‘watertight’, that is have no holes in
processes uploaded aerial imagery and can be them, and models derived from photogrammetry
used to disseminate the results. often require a considerable amount of editing
to optimise the mesh and remove, for example,
Textured 3D models can be useful under some intersecting faces, non-manifold faces, small
circumstances. They can provide ‘virtual access’ tunnels and connected components. In most
to areas that cannot normally be reached (Figure objects with complex undercutting geometry
87 shows the painted interior of Longhtorpe there will be areas that could not be covered
Tower, Cambridgeshire, to which disabled access in the survey, and some interpolation will
is compromised), can be used for generating be necessary to provide a watertight model.
sections and profiles, and can be used as the There are many pieces of software available
basis for 3D-printed physical models. commercially that can provide this functionality,

Figure 87
Photogrammetric model used to enable ‘virtual access’.

< < Contents 78

but there are also several free packages that will been obtained by using whatever was available
enable you to produce a watertight model, and for sites that have now been lost (for example the
check a model for its suitability for 3D printing. photogrammetric reconstruction of the Bamiyan
Netfabb is one example, and checking can also Buddha statues (Grun et al 2004).When using
be carried out with the open-source Meshlab. whatever is currently available, be aware that
people have a tendency to take (more or less) the
same images or views of a subject, so complete
3.7 Community involvement coverage is typically not achievable, and you
may not know whether arbitrary cropping and
The ubiquity of cameras and the high quality treatment of images with ‘special effects’ has
of images that can be achieved using, in many taken place, which will heavily compromise their
circumstances, just the automatic settings, means photogrammetric usefulness.
that projects with community (non-professional)
involvement are a distinct possibility. Early
examples of this, employing older software but 3.8 Metadata and archive
applying the same principles outlined here,
include NADRAP, which was able to generate The ADS provides several useful guides
excellent and useful results. More information regarding the metadata required for archiving
can be found on the Archaeology Data Service archaeological projects and their constituent
(ADS) website. There are many other, more components on their website. These include
contemporary, examples of community inclusion close-range photogrammetry and close-range
projects (for example McCarthy 2014 and the photogrammetry and SUA survey.
ACCORD project where photogrammetry has been
used successfully. The availability of free software In some circumstances archiving the original
(see Software (page 17)) enables the production images and control data captured on site may
of 3D models and outputs that can be scaled and be sufficient: the analytical results are usually
orientated later, provided that the scale bars and presented as part of the final report. If the ‘raw
other necessary information have been included data’ are archived correctly, the reconstruction
in the images. can always be processed at a later date, as was
often done with, for example, ante-disaster stereo
Crowd sourcing can be a good way of collecting photogrammetric imagery in the past, and the
raw imagery but it is important to ensure that user will often therefore be able to take advantage
those providing the photography understand the of subsequent developments in software and
sort of images required and that they must not hardware. This represents a significant difference
have been cropped or had special effects applied. and potential advantage compared with most
That said, some very impressive results have other types of data gathered on any site.

< < Contents 79

4 Case Studies

Case Study 1: Using archive from motion (SfM) software in comparison with
aerial imagery high-end alternatives.
Comparing digital surface models created from
archive aerial photography using SfM and However, SfM software is optimised for use
stereo-matching photogrammetry software with a large number of overlapping, converging
at Eggardon Hillfort, Dorset. images, which does not necessarily match the
requirements for SAPs, which are a 60 per cent
Introduction forward and a 20–30 per cent side overlap with
Archive stereo-aerial photographs (SAPs) parallel, or very slightly converging, geometry.
from 1948 and 1984 of Eggardon Hillfort and As an example, Agisoft LLC (2014) illustrate
its landscape, near Bridport, Dorset, were ‘capture scenarios’ in their manual and suggest
processed using photogrammetric software to overlaps of +80 per cent forward and 60 per cent
create digital surface models (DSMs). DSMs are side, which is unlikely to be met by archived SAPs.
useful for illustrative and analytical purposes in This raises the question of how well SfM software
archaeology, and their production promises to be can produce DSMs compared with outputs from
faster and cheaper with the advent of structure high-end alternatives.

Metres Metres

Figure CS1.1
The location of GCPs within the hillfort landscape.
© Heather Papworth

< < Contents 80

Software and methods bundle block adjustment to obtain an optimal
DSMs were created using SocetGXP, a high- solution. This is gauged in both packages by
end software package from BAE Systems, and examining residual errors for the ground control
PhotoScan SfM software, from AgiSoft. No camera and check points, which are given in metres and
calibration details were available for the SAPs, pixels. To refine errors, further alterations can
which were digitised from original negatives using be made to loosen or constrain the accuracy of
a photogrammetric scanner held by the Historic the camera, GCPs and check point locations, for
England Archive in Swindon. Ground control example. The process of block bundle adjustment
points (GCPs) were collected using a Leica Viva can then be repeated and the residual errors
global navigation satellite system (GNSS) re-examined. After completing this process in
(Figure CS1.1). PhotoScan and SocetGXP, the point cloud was
exported for interpolation in ArcGIS 10.1, using
The processing workflows in each software the Natural Neighbour function, to produce a
package are significantly different to each other, raster DSM with a 1m pixel resolution. The results
although both require iterative processes of are shown in Figure CS1.2.
checking and re-checking the quality of the photo

Metres

Figure CS1.2
DSMs generated from 1948 and 1984 aerial
photography using PhotoScan (left) and SocetGXP
(right).
© Heather Papworth

< < Contents 81

 dataset. This type of assessment may be sufficient
if a DSM is intended for illustrative purposes,
but data quality is an important consideration
when analysis is to be undertaken, such as the
construction of viewsheds or cost-path analysis.

Histograms illustrating the residual difference

between the GNSS and DSM elevations (Figure
CS1.3) were useful for visualising the range
of residual errors (x-axis) and how often they
occurred (y-axis). If errors within the data are
random, the shape of the curve fitted to the
graphs should be bell-shaped. If a bell curve is
not apparent, this could indicate a systematic
error present in the data, as shown by the bottom
right-hand image in Figure CS1.3. A further issue
highlighted in Figure CS1.3 was the larger range
of errors contained within both of the PhotoScan
DSMs compared with those generated using
SocetGXP. This was confirmed by the range
variable, shown in Table CS1.1, which could
indicate that the PhotoScan DSMs were not as
accurate as those from SocetGXP.
Figure CS1.3
Residual histograms showing the range and magnitude However, further analysis of the statistics
of elevation error when comparing the DSMs with GNSS
contained within Table CS1.1, such as the
elevation values.
© Heather Papworth standard deviation (SD), RMSE and confidence
intervals, suggested that the 1984 PhotoScan DSM
was slightly more accurate than the SocetGXP
DSM: the residual values were all slightly smaller
To assess the accuracy of elevation values in each for the 1984 PhotoScan dataset. For example,
DSM, an independent data set was created by the RMSE value comparing GNSS elevations with
using the Leica Viva differential GNSS to record those from the 1984 PhotoScan DSM was 0.587m
a large number of elevation values across the compared with 0.622m, the value returned from
hillfort at random locations with an uncertainly the same comparison with the 1984 SocetGXP
of less than 0.014m. These points were imported DSM. The opposite was true for the 1948 DSM
into ArcGIS 10.1 and used to extract coinciding results: SocetGXP considerably out-performed
elevation values from each DSM. Summary PhotoScan. Overall, the results demonstrated how
statistics comparing the GNSS elevations to those important it is not to rely on just one method for
of each DSM were created using the statistical assessing data quality.
software SPSS.
Conclusion
Comparison of DSMs The results of this study demonstrate that stereo-
A visual assessment of the DSMs (Figure CS1.2) matching photogrammetry software, namely
produced by both PhotoScan and SocetGXP SocetGXP, generates a more accurate DSM when
showed minimal differences between the 1984 processing older SAPs. However, DSMs produced
results, although the PhotoScan DSM contained from more modern SAPs using SfM software,
more details along the hedgerows. Despite the particularly PhotoScan, are comparable with,
1948 DSMs both containing excessive noise, the if not slightly more accurate than, those created
hillfort ramparts were visible in the SocetGXP with specialist packages.

< < Contents 82

Case Study 2: SUA imagery on landscaping associated with the construction
single sites of a stately home on the site in circa 1607 for Sir
Assessing the value of photogrammetrically Vincent Skinner and the sites of a number of 19th­
processed imagery from SUA at Thornton Abbey and 20th-century archaeological excavations.
and Ashnott lead mine. The entire precinct is designated as a scheduled
monument (List entry Number 1011198), while
Thornton Abbey the gatehouse and ruins of the claustral buildings
Thornton Abbey is the site of a major Augustinian have been in state guardianship since 1938. The
house located in North Lincolnshire at national existence of the archaeological survey was the
grid reference (NGR) TA11801895. Now chiefly main reason why the photogrammetric survey was
known for the surviving remains of the large and undertaken: to benchmark the technique against
ornate fortified gatehouse, it was the subject an existing survey and to examine the correlation
of historical, archaeological and architectural between the two, providing an estimation of the
research undertaken by English Heritage between performance of the photogrammetric product and
2007 and 2010 (English Heritage 2010b). Part of evaluating its potential contribution to the field-
this research involved a detailed archaeological recording workflow. The area flown corresponded
survey of earthworks between the claustral to the area that had been studied and presented
buildings and the gatehouse (Figure CS2.1). in the 2010 report, and measured approximately
The survey provided evidence of the medieval 500×250m.
construction sequence, as well as garden

Figure CS2.1
Results of archaeological earthwork survey.

< < Contents 83

The site was flown using a subcontractor (Skyline imminence of summer opening to the public
Images Ltd) using a Droidworx XM8 octocopter, precluded the use of paint marks, and the area
carrying a Canon EOS 5D mark III digital single of interest had no points of hard detail that were
lens reflex (DSLR) camera. Ground sample suitable as GCPs. The GCPs were used to optimise
distance (GSD; the distance on the ground the alignment of the aerial images and to put
represented by a single pixel in the images) was the survey ‘on the map’ in the correct position
specified at 40mm, which, with the lens used, and orientation. The imagery was processed
dictated flying at the legal ceiling of 120m (as using Agisoft Photoscan Pro. After structure from
stated in Civil Aviation Authority regulations). The motion (SfM) alignment and filtering, half of the
site was covered by 56 vertical or near vertical GCPs were added as control and for alignment
shots. The brief specified a front-to-back overlap optimisation, and a sparse point cloud of
of at least 80 per cent and a side lap between approximately 250,000 tie points was produced.
swaths of at least 60 per cent. A small number of The remainder of the GCPs served as check
oblique images was also included for processing. points. A dense point cloud was then generated.
Immediately prior to image acquisition, a network The camera positions can be seen above the
of control points was established across the site. point cloud in Figure CS2.2, and small flags
The ground control points (GCPs) themselves denote the GCPs.
were paper plates pinned by survey pegs: these
are cheap, clearly visible and unambiguous in The dense point cloud was classified to remove
the aerial photography and are easy to place trees, scrub and larger buildings such as the
and remove. The presence of livestock and the gatehouse and Abbot’s Lodge farmhouse (Figure
CS2.3). A mesh was generated using only those
points classified as ground. The resulting digital
elevation model (DEM) was exported for use
in a geographical information system (GIS) for
processing and analysis, along with a composite
ortho-image derived from all the input images.

Analysis was conducted using ArcGIS, although

the techniques employed were straightforward
and could be replicated in many open-source
alternatives. The DEM provided the basis for
a composite hillshade (Figure CS2.4), which
elucidated many of the ground variations hinted
at in the raw DEM. Archaeological features
were clearly visible between the gatehouse and
claustral range. In addition, a slope analysis was
used (Figure CS2.5); flatter areas are shown in
green, intermediate slopes in yellows and steeper
slopes in reds. Clearly, many other analytical
processes could have been applied to the
data. Both outputs demonstrated a very strong
correlation with the hachured plan derived from
the previous site survey (Figure CS2.6), which was
Figure CS2.2 (top) the hoped-for result, and demonstrated that, in
The dense point cloud generated from the aerial theory at least, the technique has potential for
photography.
archaeological landscape investigations. Although
Figure CS2.3 (bottom) the analyses were clearly not interpretative
The classified point cloud, showing filtered buildings, products they did show some features and areas
trees, scrub and dwarf walls.
that may benefit from further examination.

< < Contents 84

Figure CS2.4 (top)
A hillshade of the unfiltered digital elevation model.
Figure CS2.5 (middle)
Slope analysis of the digital elevation model.
Figure CS2.6 (bottom)
Extract of hachure plan overlaid on slope analysis,
showing strong correlation between the two.

< < Contents 85

Ashnott lead mine had developed and to ensure that new fences,
Based on the encouraging results from Thornton intended to safeguard the remains by improving
Abbey, it was decided to see whether the stock management, were correctly placed around
approach could be developed further to help the the most significant parts of the site. It was also
English Heritage Assessment Team with a new designed to highlight places where the collapse
request for an analytical survey of a Heritage at of old, poorly sealed shafts presented a danger to
Risk site at Ashnott in Lancashire. The scheduled livestock and hill-walkers.
remains of Ashnott lead mine lie on and within a
small limestone knoll on the edge of the Hodder The site was flown in a similar fashion to Thornton
valley north of Clitheroe, in the southern part of Abbey, this time by Aerovision UK Ltd using a
the Forest of Bowland area of outstanding beauty fixed-wing senseFly eBee small unmanned aircraft
(AONB). Documentary research has shown that (SUA) carrying a Canon Ixus ‘point-and-shoot’
the mine may have been active around 1300 and camera. Once again a network of highly visible
was certainly a going concern when Thomas markers was positioned across the target area
Proctor entered into a 3-year lease with the to provide ground control. The imagery was
Duchy of Lancaster to ‘digge, take & myne leade’ processed as above, but in this case the digital
at ‘Asshe Notte’ in 1538. By the time the mine models (for example Figure CS2.7) were used
closed in the 1830s, the victim of a general slump by the Assessment Team to draft, in AutoCAD,
in lead prices, generations of miners had created an outline plan of the earthworks similar to
a tightly knit complex of surface workings and those normally created through earthwork
underground levels chasing the erratic patterns survey, by marking lines along the tops and
of mineralisation throughout the knoll. The bottoms of slopes. This plan was then taken
survey was required to understand how this mine back into the field, where it was verified, refined
and augmented by close observation and the
judicious use of survey-grade global navigation
satellite system (GNSS) equipment. The resulting
earthwork plan, with slopes expressed with
hachures in a readily readable form (Figure CS2.8),
was somewhat less detailed than that normally
produced by traditional ground-based survey. It
was metrically accurate, however, and sufficiently
nuanced to support the archaeological analysis of
the site as described in the accompanying survey
report. Crucially, this approach was perfectly
adequate to identify the concerns that had led
to the site’s inclusion on the Heritage at Risk
Register and to provide the details required to
guide conservation measures in a forthcoming
Higher Level Stewardship agreement. Measured
against the scale of survey standards published
by English Heritage, in which Level 2 records the
general form of a monument and Level 3 captures
its complexity, this SfM-derived method sits at
about 2.5 or perhaps a little higher. It certainly
suited the requirements at Ashnott and, in terms
of the fieldwork, took less than half the time
required for a comparable level of detailed field
survey. However, while detailed and highly flexible
Figure CS2.7 three-dimensional (3D) imagery is a tremendous
Hillshade of unfiltered DEM of Ashnott Mine. tool, interpretations derived from it must still

< < Contents 86

 Case Study 3: SUA landscape survey
The 2015 Rievaulx Abbey landscape survey,
North Yorkshire.

Introduction
Rievaulx Abbey is located on the southern fringes
of the North York Moors in the Rye valley, 4 miles
upstream and west of the historic market town of
Helmsley, North Yorkshire). The abbey buildings
occupy a naturally elevated terrace to the eastern
side of a steep-sided wooded valley. The valley
floor is predominantly flat and under pasture,
housing the village of Rievaulx. The abbey remains
and their immediate surroundings are in the care
of English Heritage, while the rest of the valley is
privately owned.

The most comprehensive survey of the abbey’s

landscape has been Caroline Atkins’ 1996
investigation, which formed part of Fergusson
and Harrison’s study of the site (Fergusson
Figure CS2.8 and Harrison 1999). English Heritage’s 2001
Extract from the earthwork interpretation plan conservation plan identified the need for a full
for Ashnott, derived from the DTM and
programme of landscape survey and analysis
ground observation.
of the site (English Heritage 2001, 64). In 2005
English Heritage undertook a study of the abbey
and its landscape (Dunn and Pearson 2005).
At that time it was not possible to undertake
be informed by an experienced eye if they are to a detailed survey, so Atkins’ 1996 survey was
be robust. From the surveyor’s perspective, the utilised for the 2005 study; the published survey
most valuable parts of the process were the site report (Dunn and Pearson 2005) then represented
visit before the flight, which provided a good the most up-to-date interpretation of Rievaulx’s
understanding of the site prior to mapping the landscape. As this survey is now more than 20
patterns observed from the air, and the detailed years old, the site has recently been re-surveyed
reassessment of the SfM-derived plan when it using modern techniques.
was taken back to the site. Only then did the
finer distinctions between paths and Methodology
watercourses, washing floors and working areas, The survey was conducted using a structure
become fully apparent. from motion (SfM) and multi-view stereo (MVS)
approach, using low-level aerial photographs. The
Rievaulx landscape provided the opportunity to
test this methodology against traditional survey
techniques and Environment Agency (EA) 0.5m 2
lidar data. A circa 48ha survey area was identified,
which encompassed the abbey remains, the
village and surrounding farmland. Historic
England commissioned an external contractor
to overfly the area using a small unmanned
aircraft (SUA). This captured overlapping vertical
photographs from an altitude of circa 180m,

< < Contents 87

 operated in shutter priority mode, ensuring a
fast shutter speed was maintained, improving
the chances of capturing clear images; a speed
of 1/800th of a second and an ISO of 800 were
used. The f/stop varied between f/2.8 and f/5.6 to
compensate for the variable lighting conditions;
Figure CS3.2 shows an example image.

Images were processed using Agisoft Photoscan

Professional. The SUA on-board software logged
the camera’s position and orientation during
each image capture. This combined with the
images was uploaded for processing. Initial image
alignment used the SUA log, and then this was
refined using the GCPs, which were manually
identified in each image. Using the SUA log
considerably decreased the processing time for
the initial image alignment. The SfM approach
created a continuous ortho-image, then a DSM
0 200m was generated using the MVS process. The DSM
was accurate to circa 20mm in the horizontal and
circa 35mm in the vertical plane. This accuracy
was measured using check points that were
GCP that had not been used for refinement.
Once generated, the DSM was manipulated
Figure CS3.1 in a geographical information system (GIS) to
Hachure plan derived from the new survey. assist the identification and interpretation of
topographical features.

Results
80m above the valley floor. To facilitate Following the GIS manipulation, a range of
georeferencing, ground control points (GCP) features was identified. In some cases the
were installed across the valley using a survey- survey enhanced existing knowledge, and in
grade global navigation satellite system (GNSS), others it identified previously unknown features.
employing markers identifiable in the images. These features were visualised as a hachure plan
(Figure CS3.1) to allow direct comparison with
The SUA was equipped with a Sony Alpha the 1996 survey. An example of this enhancement
ILCE-A6000, a 24.3MP digital camera, triggered could be seen when the southern precinct wall
by the SUA on-board software. The automation was examined. The SUA data showed this as a
facilitated a pre-determined ground sample low bank, curving as it crossed the valley; it was
distance (GSD) and level of image overlap of apparent that this curve followed that of a relict
0.05m and 80 per cent, respectively. To ensure river channel (Figure CS3.3). The channel may
total coverage, two flights were undertaken with have been visible as a shallow boggy depression
identical flight plans, although the actual paths when the precinct wall was constructed and
differed slightly because of localised weather offers an explanation for the curving bank;
conditions. Six hundred and six images were the monks may have chosen to enhance the
captured. Each image was manually quality precinct wall using this natural feature. This
checked to ensure only clear images were relationship had not been previously identified
processed; 42 images were removed from the and without the SUA methodology it would have
data set because of their poor quality. The camera remained unnoticed.

< < Contents 88

Figure CS3.2
Example of the imagery captured using the SUA.

Evaluation
This case study highlights the nature of survey
available via the SUA methodology. It is high
resolution and highly accurate, although there
were significant areas where no data could be
obtained because of vegetation coverage. A lidar
approach delivers greater coverage at a lower
resolution, which can be seen in the comparison
of the methodologies illustrated in Figure CS3.4.
Nonetheless, these methodologies should be
viewed not as competing but as complementary.
Ideally SUA data should be obtained for open
areas and lidar data for wooded regions, thus
optimising coverage and resolution. The SUA
approach allowed data to be collected from a
large area quickly; the initial data collection
was achieved in half a day and the subsequent
ground-truthing in one day. An alternative
methodology is ground-based survey, but for this Figure CS3.3 (top)
scale of survey a team would need to be deployed Hillshaded view of the southern outer precinct wall
showing the curving bank following the line of the
for an extended period of time to collect the
relict river channel to its south.
requisite data, and it is questionable whether the
same resolution could be achieved. Figure CS3.4 (bottom)
Comparison between SUA (left) and EA lidar (right)
imagery.

< < Contents 89

Not all locations are suitable for the SUA terrestrial 3D laser scanning was used to record
methodology; areas with significant tree cover the tunnels, which could not be surveyed from
will not be productive. Locations with adverse the air.
weather conditions are not ideal for SUA either.
Even at Rievaulx, in a relatively amenable Historic England subcontracted the acquisition
location, poor weather caused the survey to be of the aerial imagery to Future Aerial Innovations.
rescheduled on three occasions. However, the Given the nature of the terrain, it was decided that
SUA approach drastically reduced the fieldwork the most efficient coverage would be obtained
required, as the majority of the work was desk- using a combination of fixed-wing and multi-rotor
based. This reduced staffing costs compared platforms to capture both vertical and oblique
with traditional methodologies, although the imagery. While the vertical imagery from a fixed-
data-capture and processing costs do need to be wing platform is perfect for larger 2.5D terrain
taken into consideration. Nonetheless, models, this project required the cliff faces and
compared with ground-based survey, the SUA castle ruins to be as detailed as the rest of the
methodology can deliver savings and should be terrain to create a full 3D model. The fixed-wing
considered at suitable locations. The resulting platform, using a camera taking vertical imagery,
plan has added detail to the 1996 survey and was used for efficient coverage of the upper
thus shed new light on the Rievaulx landscape. surface of the site, while the multi-rotor system
This survey will ultimately help inform continued was used for the cliffs and other vertical elements,
management of this significant site and help such as the walls of the structures on the site.
protect it for the future. The fixed-wing solution could be flown on a pre-
planned flight path, while the multi-rotor was
flown manually. Although the entire site could
Case Study 4: Combining SUA and theoretically have been captured using a multi-
terrestrial data rotor small unmanned aircraft (SUA), the use of
Tintagel Castle, Cornwall. a fixed-wing SUA represented considerable time
savings given the area to be surveyed.
Introduction
In late 2014, work was undertaken by Historic Constraints
England on behalf of English Heritage to produce The subcontracted aerial work, given project
data for use in a new exhibition at Tintagel Castle, deadlines, had to be flown in November and
on the north coast of Cornwall. The brief included December of 2014. The location of Tintagel is
the production of a model of the site suitable such that, at that time of the year, there are strong
for three-dimensional (3D) printing and other prevailing winds blowing in from the Atlantic,
potential uses, including archaeological analysis lighting is challenging, and weather conditions are
and illustration. very variable. This meant that the multi-rotor SUA
(an Ascending Technologies Falcon 8 equipped
The area to be surveyed comprised circa 2.5km 2 with a Sony A7R camera), which can be flown
of highly variable terrain, including the castle in windier conditions, was used until weather
and associated structures, two tunnels (one windows appeared when the fixed-wing SUA
natural and one artificial), sea cliffs, and several (a senseFly eBee, equipped with a Canon
other areas inaccessible from the ground. The Powershot S110) could be deployed. Capturing
accuracy requirements for the project involved the entire data set took several visits to the site.
the production of the 3D data with a ground
sample distance (GSD) of approximately 40mm. There were also constraints on the terrestrial
In order to achieve this, photogrammetric survey scanning. Merlin’s Cave is tidal, so only relatively
from an aerial platform was the obvious choice for short windows of opportunity were available for
those parts of the island visible from the air, while the survey work there.

< < Contents 90

Flight operations
The camera on the multi-rotor was equipped
with a 35mm prime lens. While the multi-rotor
was capable of waypoint-driven flying, on the
mainland valley area it proved to be more
effective to fly it in manual mode, to give the
camera operator time to frame images correctly.
When flying in manual mode, extra care has to be
taken to ensure that sufficient overlap is present.

Flying in waypoint mode proved to be more

effective when capturing the oblique imagery of
the island. The flight plan set the SUA 100m out to
sea, at a height level with the top of the island and
looking back towards the island, always focusing
on a central point. A distance of 100m ensured
the island was fully framed in one shot while
maintaining the necessary 40mm GSD. To keep
the SUA in sight and within 500m, both camera
operator and pilot followed its path around the
island on foot.

Considerable testing was required to get the

correct flight height, shutter speeds and distances,
especially in the tough weather conditions. Once
fully set up, a total of 15 flights was required to
complete the oblique phase of image capture.

A follow-up visit during more favourable weather

was needed to capture the fixed-wing imagery,
which proved to be a relatively simple task using
the waypoint-driven capabilities of the SUA. Four Figure CS4.1 (top)
flights were required to provide vertical imagery Arrangement of the images used.
in two directions to ensure maximum coverage Figure CS4.2 (above)
and overlap (Figure CS4.1). The standard camera Ground control points (in yellow) used for the survey.
was swapped for the upgraded Canon Powershot
S110 with shutter priority mode. Given the
prevailing conditions, without this the images
would have been blurred. using a survey-grade global navigation satellite
system (GNSS) to an accuracy of approximately
Accurate ground control is essential in a project 10mm in x, y and z.
with so much topographic variation. A total of 25
ground control points (GCPs) was used (Figure Laser scanning
CS4.2). Removable markers were specified, so Terrestrial laser scanning was undertaken in
that no trace would be left once the survey was two areas that could not be covered by the
completed. Because of the steep cliffs and rugged SUA imagery. These comprised the artificial
ground, putting in the ground control took a full tunnel on the top of the island, the use of which
day. The position of each point was surveyed remains open to interpretation, and the natural

< < Contents 91

 Merlin’s Cave. For both areas, a Faro Focus 3D
laser scanner was used. The main consideration
was portability. The tidal Merlin’s cave required
the survey be undertaken quickly, and the very
variable conditions underfoot meant that a light
instrument could only be used safely with two
operators, keeping a watchful eye on the tides at
all times. The laser scanning was also controlled
by survey-grade GNSS, ensuring that the data
could be integrated with that generated by the
aerial photogrammetry.

Processing
The terrestrial laser scanning was undertaken
and processed by the Geospatial Imaging Team
at Historic England using Faro Scene. The number
of scanner setups and control is shown in
Figure CS4.3.

Processing of the aerial data was also undertaken

by Historic England. A total of 662 images, both
oblique and vertical, was supplied in RAW and
TIFF formats along with the GCP positions and
coordinate schedule. With the GNSS GCPs added,
the data processed to an overall positional
Figure CS4.3 accuracy of 0.02m in x, 0.04m in y and 0.03m
Scanner setups for surveying Merlin’s Cave. in z (RMSE). Data processing was undertaken
using Agisoft Photoscan.

Figure CS4.4 Figure CS4.5

Noisy inputs from a point-and-shoot camera. Cleaner images from a better camera.

< < Contents 92

Some issues became apparent at a fairly early
stage of processing the vertical imagery. Given
the payload constraints of a fixed-wing solution,
only a relatively compact camera could be used.
Because of the prevailing weather conditions,
with high winds on overcast days, the camera
was liable to increase the International Standard
Organisation (ISO) values dramatically in order
to obtain well-exposed images, given the high
shutter speeds necessary to avoid motion blur.
This had the knock-on effect of introducing Figure CS4.6
considerable noise into the imagery (Figure Noisy inputs for the range of buildings on the left
resulted in noisy data in that area of the model.
CS4.4). Using noisy inputs meant that the model
lacked clarity in some areas, and Historic England
requested some areas to be re-flown with the
multi-rotor equipped with a much better camera
to achieve sharper images. When these images The terrestrial laser scanner and the
were received the difference in quality was photogrammetric data were combined and edited
considerable (Figure CS4.5). in Geomagic Wrap. This also involved conditioning
the data so that is was suitable for submission to
Unfortunately, scaffolding had been erected in the 3D printing company, such as removing the
the intervening period between flights, so only self-intersections, small holes and small tunnels
the noisier imagery could be used for the digital that are often produced during the meshing
reconstruction of the buildings housing the site’s process and ensuring that a clean, ‘watertight’
tea rooms (Figure CS4.6). Despite Figure CS4.6 mesh could be produced. The model was then
showing a draft model processed at relatively re-imported into Photoscan for texturing and final
low resolution to illustrate the problem, the export at a variety of resolutions.
difference in reconstruction quality between the
two areas is striking, and clearly demonstrates Products
the issues associated with the use of high ISO Given the necessary outputs for 3D printing and
values on cheaper cameras with smaller sensors. reconstruction drawings, the highest resolution
The higher resolution imagery also helped model generated for the whole area comprised 42
considerably with the definition of archaeological million polygons, although, after trying the model
features such as low walls, which were formerly at a number of different resolutions and texture
more difficult to distinguish from the thick tufts of sizes, the final one used for printing comprised
grass surrounding them. approximately 15 million polygons. This was then
used to construct the display model, originally
Other issues were also generated by using routered from hardened foam. The net result was
imagery captured at different times. One of the an accurate scaled model of the island and a
products was a low-resolution textured 3D digital section of the adjoining mainland (Figure CS4.8).
model to be used by graphic artists The model used by the reconstruction artist
subcontracted to English Heritage to help with comprised approximately 2 million polygons, as
reconstruction drawings. As the original image with a high-quality texture this retained enough
set had been captured at low tide and the geometric detail for its purpose.
second set at high tide, considerable masking of
images had to be done to ensure that the final
texture represented the low-tide ground surface
accurately (Figure CS4.7).

< < Contents 93

 Figure CS4.8
The model of the island in the visitors’ centre.
© English Heritage

The printed model now forms part of a display

in the new visitors’ centre at Tintagel, with an
overhead projection system overlaying the
model with video showing the development of
settlement and use of the island, in conjunction
with an audio soundscape. The reconstruction
drawings, based in part on the model, also form
part of the exhibition (Figure CS4.9).
Figure CS4.7
Texturing problems resulting from flights at
different times.

< < Contents 94

Figure CS4.9
One of the reconstruction drawings, depicting Tintagel
in the Dark Ages.
Image by Monumental © English Heritage

Case Study 5: Terrestrial imaging on re-visited in a way not otherwise possible. Some
archaeological sites enthusiastic promoters of the method argue,
falsely, that it can be used to ‘preserve heritage’:
One of the principal objectives of all an absurd assertion. What it can do, however,
archaeological recording is to create a record is accurately map a 3D surface and drape high-
that is sensitive to post-excavation examination resolution photographic textures on that surface
and re-interpretation. It is this requirement in such a way as to facilitate interaction and close
that makes the use of photogrammetry so examination that may not even be possible in
revolutionary, as accurately created three- the field. Three examples are presented here of
dimensional (3D) models, at appropriate the application of photogrammetric methods to
resolutions, allow excavation data to be archaeological excavation.

< < Contents 95

Roulston Scar, North Yorkshire to allow for later re-assembly in the archive, was
During the winter of 2013–14 an excavation was in this case impractical; the ditch was so large that
carried out to examine the defensive rampart the section had to be extended largely by machine
and ditch on the inland side of the promontory to comply with health and safety requirements.
fort at Roulston Scar in the North York Moors
National Park, North Yorkshire. The excavation The primary trench was only 1.8m wide, so targets
was undertaken to try and recover dating and to provide a sufficient density of accurate points
environmental evidence and learn more about for georeferencing the 3D model were fixed at 1m
the scale of the defences on the eastern side intervals on the surface just beyond the edges of
of the hillfort, where they run along the steep the trench. Additional targets were installed in
side of the promontory. The excavation was the base of the trench and in the sections; these
undertaken to document a single section through were required to give a more 3D georeferencing
the rampart and ditch. It was intended to be a point network that covered the trench both in
2-week project but it was quickly realised that plan and from a sectional point of view (Figure
the scale of the monument was such that this CS5.1). Although a single network of points placed
would be impossible. The trench was located at around the edges of the trench would doubtless
the point where the natural topography of the have supported the production of a high precision
hill created a very small ditch cut into the natural plan or ortho-image, to maintain the same
slope that would have served as an effective precision vertically within the trench other points
defence. However, the ditch in fact measured were necessary along the base and sides of the
6m wide and over 2m deep, cut into bedrock, trench. The targets were printed on high-density
and was too deep to excavate in the original 2m matt plastic, with industrial-grade hook-and-loop
wide trench and had to be extended to maintain tape mounted on the back and firmly attached
safe working conditions. The prevailing weather to perforated mounts nailed in place with
conditions were also poor, so it was decided at 100–150mm nails. This approach is flexible, the
the outset to record the plan and sections using targets are washable and reusable, and it is easy
photogrammetry for the primary record, backed to position them even when ground conditions
up with interpretive drawings. The objective to make it difficult to find a secure mounting point
record each deposit individually prior to removal, at first attempt. To ensure accuracy, where targets

Figure CS5.1
A completed 3D model showing the excavated section
through the rampart and ditch at Roulston Scar, viewed
in Sketchfab. Note the positions of the georeferencing
markers around and inside the trench.
© Dominic Powlesland

< < Contents 96

are not firmly attached to static structures, they Figure CS5.2 shows a screen-shot in Agisoft
should be re-surveyed each time a new model Photoscan Pro of a model of the primary trench
is generated. through the defences at Roulston Scar, with the
camera positions marked in blue. A photo mast
It is essential that every part of the subject to was set at an elevation of about 4m, and two
be modelled is covered by a minimum of three strips of images recorded along each side of the
overlapping images, ideally from a position trench. A third strip was taken using a tripod
that is at a right angle to the area to be modelled. within the trench to record the section, in each
This requires careful observation in case the camera was moved at approximately
environments where stone structures, etc, 1m intervals. The ditch was recorded using extra
have considerable depth. high-level pole photographs and further runs of

Figure CS5.2 (top) Figure CS5.3 (bottom)

A screen-shot from Agisoft Photoscan Pro showing A screen-shot covering an oblique view of the north-
a vertical view of the excavation trench with camera facing section through the hillfort rampart showing
positions in blue. camera positions for recording plan and section.
© Dominic Powlesland © Dominic Powlesland

< < Contents 97

 images facing each section. Figure CS5.3 shows
an oblique view of the excavated trench through
the modelled rampart at Roulston Scar, showing
the camera positions used to provide coverage for
a complete model reconstruction, ensuring that
detail can be viewed in plan and in section.

It is important when working in the UK to

appreciate that when georeferencing using
national grid coordinates recorded to millimetre
precision, as generated when saved from the
survey instrument, the values are too large to
be properly handled within most 3D computing
environments, which rely upon the use of Open
GL software. To avoid such problems coordinates
should be reduced to exclude the 100km and
10km values at the beginning of each easting or
northing value, respectively. If by unfortunate
accident you happen to encounter coordinates
that cross the 10km or 100km boundaries, a site
grid should be used with an identified offset to
reduce the number of digits in each axis. Any such
shift should be documented and recorded as part
of a log file describing the processes and software
used, and added to the archive.

Once a model has been created in the field,

it should be checked for completeness, scale
and accuracy and checked against the physical
evidence to underpin the overall recording
process. Publication, giving live access to the 3D
models, can be achieved using Sketchfab and
other online tools; the data can be archived as
an object model with its texture files as well as
in Adobe 3D PDF format. Agisoft Photoscan Pro
and many other software packages can be used
Figure CS5.4 (top) to create 3D PDF files; experimentation using
Completed 3D model incorporating an annotation and different software has shown that the output from
archived as a 3D PDF file.
the same 3D model is not always the same, but in
Figure CS5.5 (middle) the long term this is not critical if the source files
A view of the completed 3D model of the excavated are archived with the PDF and the export software
rampart section, published as a scaled model in PDF
used is recorded. 3D PDF files are internationally
format suitable for accurate measurement purposes.
accepted as a recognised archive format and, if
Figure CS5.6 (bottom) correctly scaled, give the user unprecedented
Excavation photograph showing a late palisade trench access to the model in ways that can help
cut behind the rampart of the prehistoric hillfort. The
support future research. Figures CS5.4 and CS5.5
stratigraphy suggests that the palisade post-dates the
fully eroded rampart and could relate to the Battle of are screen-shots showing the Roulston Scar
Byland in 1322. rampart section annotated with observed notes
© Dominic Powlesland and measurements.

< < Contents 98

The Roulston Scar excavation was inspired by a
need to assess the date of this monument and
secure environmental evidence that would allow
it to be placed in the wider environmental context
emerging from the examination of peat and pollen
evidence recovered from the very much smaller
hillfort at Boltby Scar. The use of photogrammetry
produced a far higher quality record than would
otherwise have been possible, given very tight
time constraints and poor winter weather. In this
case the very limited amount of time meant it was
not possible to process the models and examine
them on site before the trench was backfilled.
Careful examination of the 3D models after the
excavation was complete revealed many details
not observed in the field and a further small
excavation was undertaken in January/February
2015 to try and resolve the questions arising from
the investigation of the model and the need to
secure dating evidence for what appeared to be
a very much later palisade trench behind the
rampart (Figure CS5.6).

The advantage of correctly georeferenced 3D

models is clear when the 3D models from the
two excavations at Roulston Scar are combined.
The models were loaded together and the
trench edges manually clipped where the two
excavations intersected (Figure CS5.7).

Another benefit of using 3D models compiled

using photogrammetry, which may not at first be
apparent or even intuitive, is that when viewed on
screen with an object viewer such as Sketchfab
or using Agisoft Photoscan there are perceptible
differences between the front and back. Logic
suggests that this should not be the case, as the Figure CS5.7 (top)
photographic texture draped on the model is the 3D models from two trenches excavated in 2013/2014
and 2015 combined using the georeferenced data.
same; but while the applied texture is the same,
the surface upon which the texture is draped Figure CS5.8 (middle)
is different when viewed from either side. The The south-facing trench section seen from the inside
and outside. This image from a screen dumps lacks the
process of trowelling the surface, whether in plan
acuity when viewed in 3D.
or section, even when carried out to the highest
standard, leaves trowelling artefacts such as Figure CS5.9 (bottom)
smearing of the soils; in particular, where stones Detail of a section viewed from ground-level showing
a hearth, part of which is buried by collapsed rampart
become dislodged from the sections they expose
material. Without using a 3D model, this view could
parts of the section that have not been smeared only be seen by placing your head on the ground
and, when viewed from the reverse side of the surface, which is practically impossible.
© Dominic Powlesland

< < Contents 99

 Cook’s Quarry, North Yorkshire
Some archaeological sites are so fragile or have
features of such poor or very short-lived visibility
that they cannot be conventionally recorded
without, for instance, constant wetting of the soil
to maintain feature visibility; something this is not
always practical. At Cook’s Quarry, West Heslerton,
excavations undertaken ahead of sand extraction
have been in progress over many years. The sandy
soils and blown sands that are characteristic
of this multi-period site dry out almost as soon
as exposed, minimising the visibility of a large
percentage of the exposed archaeological
features. In response to the ground conditions,
excavation tends to be conducted in small
sections that progressively cover larger areas. For
much of the excavation this approach is effective,
but in some areas this makes it impossible to see
extensive areas at any one time.

Figure CS5.10 (top) Excavations in the 1980s identified a set of

Cook’s Quarry, North Yorkshire. A 3D model produced features originally interpreted as plough marks
from images captured using a SUA to map ancient
but later shown to be wheel-ruts in a hollow way
wheel ruts before drying of the sands made them
almost invisible. that was progressively filling with blown sand as
the features were formed. The slight trace of the
Figure CS5.11 (bottom) wheel ruts bounded by slightly denser sands and
An enhanced version of the ortho-image shown in slight iron-panning rapidly dried out. By using a
Figure CS5.10 to increase the visibility of individual
small unmanned aircraft (SUA) flown with a digital
wheel ruts and underlying features.
© Dominic Powlesland camera, we were able to create scaled ortho­
image covering large segments of the track-way
as they were exposed and cleaned, which will
ultimately be combined to form the basis of an
accurate plan. Because the textures used to show
section in a model, they have greater presence surface detail on the 3D model are derived from
as the holes are inverted and therefore project digital photographs, they can also be enhanced
towards the viewer (Figure CS5.8). using conventional image processing techniques
to increase contrast or change the colour balance
When viewed by eye, the section in the side of to help isolate or emphasise detail. Image
a trench can only be viewed conveniently from enhancement can be applied prior to texturing the
certain positions; using a 3D model, the viewpoint model or to the ortho-image output from a model
can be set at any position, making it possible to (Figures CS5.10 and CS5.11).
view details otherwise effectively impossible to
see. Figure CS5.9 shows where burnt soils related Star Carr, North Yorkshire
to a hearth behind the rampart at Roulston Scar The fragility of some archaeological deposits
are clearly buried by collapse from the rampart. is such that they decay on exposure, and
when exposed over large areas are difficult or
impossible to model using standard procedures.
Excavations of waterlogged deposits are one such
example: the anaerobic conditions prevalent in
waterlogged peat mean that organic materials

< < Contents 100

such as timber survive, in a way not encountered waterproof draughting film and used for recording
on dry-land sites, but as soon as these materials annotations directly in the field as the timbers
are exposed to the air they start to decay and were lifted.
distort rapidly. Although timber, for instance, can
be very well ‘preserved’, it does not necessarily There is no doubt that using photogrammetry
survive intact and unchanged; damage or decay has the potential to transform the archaeological
of parts of the internal structure combined with recording and excavation process, even if
pressures arising from the burial environment can we have to modify the traditional workflow.
reduce buried tree trunks to squashed objects Photogrammetric recording does not replace
that look like smoothed planks. Excavations traditional methods, but it allows us to combine
of waterlogged sites in peat are particularly digital modelling of excavations in progress with
challenging, not only because excavated material drawings to articulate our interpretation of the
decays and potentially distorts as it dries out, but evidence rather than serve as poor reflections of
also because the peat matrix within which the what we see, and offers the potential of a new
material is often found experiences deflection approach that by saving time at the primary
when someone stands upon or walks across it; documentation stage can free up time for more
this can affect the model-building process as the detailed and careful observations during
very material being modelled can move as it is fieldwork and thus improve the quality of the
recorded, and thus cause the resultant model excavation results.
to be imprecise with fuzzy boundaries. This is
the perfect environment in which to use an SUV
for image capture. An SUV was flown as low as a
metre above the Early Mesolithic ‘timber platform’
at Star Carr to see whether a useful addition to
the site archive could be produced. The level of
detail recovered in this experiment was affected
by the fact that the timbers had been exposed
for more than a week and, although covered and
kept wet, had suffered surface oxidation; the area
was not completely drained prior to modelling,
to minimise any risk from water flow through
the underlying deposits (Figure CS5.12). The
small sensor and fish-eye lens used in a GoPro
camera would not be the first choice to achieve Figure CS5.12
detailed high-resolution images, but the very Star Carr, North Yorkshire. A timber platform modelled
using an SUA controlled from several metres away to
low altitude at which the images were captured
reduce any impact from deformation, which could arise
meant sufficient detail was obtained to produce if collecting the images while walking between the
a high-quality model without adversely affecting fragile waterlogged timbers or close by on the side of
the site. The fish-eye lens meant that photographs the trench.
© Dominic Powlesland
taken from a vertical position at very low
elevations recorded not only the upper surfaces
of the timbers but also the sides of adjacent
pieces. The lens characteristics of the GoPro are
so well understood that Agisoft Photoscan was
able to process the images without any prior
image correction. The images for this model
were collected in a single session of less than
an hour and processed overnight to produce a
scaled ortho-image, this was then printed on

< < Contents 101

Case Study 6: Underwater early indications, laser-based technology has
photogrammetric survey yet to offer a cost-effective mainstream practical
Structure from motion: a diver-based approach methodology for underwater survey. While sonar
to photogrammetric survey. surveys are excellent for large area coverage, they
generally are not applied at the sub-metre scale
Developments in the capabilities of multi- and do not capture colour data.
image photogrammetry have been enormously
important for archaeologists, allowing practical Photogrammetry can also be used to produce
and low-cost solutions for survey of terrestrial realistic three-dimensional (3D) models
archaeological sites. However, it is within the that support dissemination of the results of
discipline of underwater archaeology that that archaeological projects to the wider public, and
photogrammetric advances have had the greatest this is an area where underwater archaeology
impact. The reasons for this are simple: the has benefited to a proportionally greater degree
logistical demands of underwater archaeology because the sites clearly are not accessible to the
in terms of time, cost, complexity and safety vast majority of the non-diving public. It is now
are much higher than they are for the terrestrial becoming possible to bridge this gap through
equivalent, and the use of photogrammetry data-driven 3D reconstruction and virtual reality.
can allow much faster and denser data capture Although 3D reconstruction has been possible
than has previously been possible. In addition, for many years, the dense and textured surface
alternative methods for the capture of similar capture of underwater photogrammetry allows
data have been far more limited for underwater communication of the visual experience of visiting
archaeology. Terrestrial laser scanning an underwater site with an authenticity previously
produces broadly similar results as multi-image impossible using the more familiar artistic
photogrammetry and has been available to and subjective reconstructions. Essentially the
archaeologists for many years; despite promising public can now see and interact with underwater

Figure CS6.1
Photogrammetric models of an intertidal shipwreck at
Ardno, Loch Fyne, Argyll and Bute.
© Wessex Archaeology, John McCarthy

< < Contents 102

archaeology in a way that is much more akin to „ the presence of moving marine life,
the actual experience of diving on the site. including fish and kelp.
Finally, the technique has proven to be highly
effective for recording intertidal archaeology, In addition to the normal requirements of
where rapid recording is essential because the photogrammetric survey (adequate coverage
sites are only accessible for a short period of time and overlap, etc) the effects of each of these
(Figure CS6.1). environment-specific factors must be carefully
considered and accounted for when planning
Maritime archaeologists are still exploring the a photogrammetric survey underwater. In one
methodologies and applications of underwater regard photogrammetry is actually made easier
photogrammetry. However, even within the last under the water: divers with good buoyancy
5 years it has been demonstrated that entire control can position themselves freely over the
shipwrecks can be recorded using this technique site, whereas ladders or small unmanned aircraft
in a single day, an output that would have (SUA) might be needed for a similar site on land.
required hundreds of hours of underwater work
setting up survey grids and using trilateration. As with any photogrammetric survey, it is
A variety of approaches have been developed necessary to be aware of the pitfalls of the
and current workflows vary from those requiring technique, particularly with regard to metric
complex custom-built rigs costing hundreds accuracy. The loss of visibility over distance
of thousands of pounds to simple diver-held when working underwater means that photos
consumer-grade compact cameras in a must be taken within a certain maximum distance
waterproof housing. from the subject (usually between 0.5 and 10m in
UK waters). Underwater surveys can accumulate
There are a number of specific challenges to error more easily over distance without the
the practice of photogrammetry underwater. possibility of correcting this using wider shots
These include: taken from further away, a technique that can
be used to help correct surveys on land. It is
„ reduced camera field of view and optical also important to realise that photogrammetry
characteristics of the lens because of air can be difficult to apply to many archaeological
to water light refraction underwater sites, because they lie in areas with
generally poor through-water visibility or because
„ low light levels they are covered in marine life that is in constant
motion. In some cases kelp can be removed, but
„ limited and varying degrees of visibility this is laborious and it may also threaten the
through the water column stability of certain sites where the kelp helps to
reduce water currents and their roots help to
„ loss of part of the colour spectrum (mainly stabilise the archaeological matrix. However,
red) in proportion to the depth plus distance when conditions are right, with clear water, good
from subject lighting and a site that is free of marine life, the
results can be astounding!
„ suspended particulates in the water column
(affecting autofocus) Two examples of archaeological photogrammetry
undertaken in the UK will be used to illustrate a
„ dappling of light near the surface (caustics) simple diver-based approach.

„ the need to avoid disturbing sediment,

particularly when photographing at
low angles

< < Contents 103

Gun Rocks, Northumberland covered in dense kelp; as the kelp was not a key
In the 1970s, divers from the Tyneside 114 factor in protecting the integrity of the site, it
British Sub Aqua Club (BSAC) discovered a large was manually cleared from the three cannons
number of cannons on the seabed at Gun Rocks, and their immediate surroundings during a
a small outcrop in the Farne Islands, off the single dive. The photogrammetric survey was
coast of Northumberland. The site is thought then undertaken in an oval pattern around each
to represent a wreck of 18th century date and cannon, maintaining a distance from camera
mainly comprises a collection of 19 cannons to subject of approximately 1m. The 3D model
lying on rocks and sand at a depth of 15m. In results (Figures CS6.2–5) proved highly detailed
summer 2013, on behalf of English Heritage and corresponded with manual measurements,
and with the assistance of Tyneside 114 BSAC, and also proved suitable for further analysis. They
Wessex Archaeology investigated the site as were later shared with ordnance experts leading,
part of the Heritage at Risk programme (Knott to a new interpretation of one of the cannons and
2013). As well as traditional survey and sonar further evidence of the possible date of the wreck
survey, the opportunity was used to undertake and the purpose of its final journey. Models were
photogrammetric survey on three of the cannons uploaded to online interactive portals to allow
as through-water visibility was excellent at over members of the public to explore the site in the
5m. The cannons and rocks around them were same way).

Figure CS6.3 (top)

Surface enhancement of a 3D model of cannons from
Gun Rock, Farne Islands, Northumberland.
© Wessex Archaeology, John McCarthy

Figure CS6.4 (bottom)

Radial surface analysis of a 3D model of cannons from
Gun Rock, Farne Islands, Northumberland.
© Wessex Archaeology, John McCarthy and Peta Knott

Figure CS6.2
Photogrammetric models of three cannons from
Gun Rock, Farne Islands, Northumberland, and one
recovered cannon on dry land.
© Wessex Archaeology, John McCarthy and Peta Knott

< < Contents 104

Drumbeg, Sutherland
The Drumbeg wreck site was discovered by
scallop divers off the north-west coast of Scotland
in the 1990s. The site was first reported and
archaeologically surveyed in 2012 (McCarthy
2012), when it was found to consist of three
cannons lying together over a section of hull
together with two anchors lying at some distance
away. The wreck has been tentatively interpreted
on typological grounds as a possible Dutch
trader of 17th century date. The site lies in an
area of open sand at a depth of around 12m and
through-water visibility is typical of the west
coast of Scotland and is often over 5m. The site Figure CS6.5
has been subject to photogrammetric survey Line drawing and volumetric calculations of
cannon 6002.
twice, first in 2012 (McCarthy et al 2015), when
© Wessex Archaeology, John McCarthy
the individual cannons were surveyed at close
range, and again in 2014, when a large area survey

Figure CS6.6
A still from an animated reconstruction of the cannons
at the Drumbeg wreck site.
© Wessex Archaeology, John McCarthy

< < Contents 105

 Figure CS6.7 (above)
Photogrammetric wide-area survey of the Drumbeg
wreck site, Sutherland, covering an area of 35m 2.
Crown Copyright Historic Environment Scotland, John McCarthy

Figure CS6.8 (left)

A simple rig for maintaining a constant distance to
the seabed.
Crown Copyright Historic Environment Scotland, John McCarthy

was undertaken. The first survey’s methodology

was largely similar to the Gun Rocks example,
but additional reconstruction of the cannons
and anchors was undertaken, drawing on the
3D survey data (McCarthy and Benjamin 2014,
Figure 6; Figure CS6.6). The second survey was
principally a sub-seabed sonar and magnetometer
investigation undertaken in order to establish
whether the remains extended further than was
visible on the seabed. At the same time a wide-
area photogrammetric survey was undertaken
in order to provide a wider context for the site,
and the results demonstrated the possibilities
for such larger surveys. In this case an area
of approximately 35m 2 was captured to sub­
centimetre detail in a single 40min dive (Figure
CS6.7). Maintenance of a constant distance to the
seabed was ensured through the use of a simple
custom rig with laser pointers rigged to a surface
buoy that intersected at a predetermined distance
(Figure CS6.8).

< < Contents 106

Case Study 7: Modelling small objects

Introduction
This case study illustrates the process of
imaging a very small object in order to provide a
metrically accurate model for three-dimensional
(3D) printing and display purposes. The object
surveyed was a Roman coin, a Denarius of Nerva
(accession number 768543), held in the English
Heritage collection at Wrest Park (Bedfordshire).
The brief was to create 3D models of some of
the finds in the Wrest Park archaeological store
that were of sufficient quality to be ‘blown up’
and 3D printed. The 3D prints were to be used for
educational visits and handling, including, for
example, the creation of rubbings. The quality
and resolution of the models needed to be such
that they would withstand an increase in size from
approximately the size of a five-pence piece to a
dinner plate. However, the purpose of the 3D print
was to produce a hardwearing replica that, while
being recognisable and accurately depicting the
source artefact, did not have to be of museum
replica quality.
Figure CS7.1
Cameras and lenses The tent and lighting setup used.
In order to image such a small object it was
decided to use macro lenses. Two cameras were
used: a Nikon D800 (36MP) with a fixed focal
length 200mm Nikkor macro lens, and a Sony Such tents are highly portable (folding down into
Alpha 7R II (42MP) with a fixed focal length Sony a small flat package that is lightweight), relatively
90mm macro lens. The Nikon setup was used for cheap and can be deployed in seconds. They are
close-up images of the coin itself, while the Sony used in conjunction with an external LED lighting
setup was used to provide slightly wider context- rig, which can be left on for the duration of the
setting shots, which included ancillary objects shoot. The light is diffused by the material of the
such as scale bars and colour reference cards. tent and produces conditions inside, where the
object is placed, suitable for taking consistently lit
Both cameras were tripod mounted, with a 5s and shadow-free images.
shutter delay to allow the camera to settle before
the image was exposed. The Nikon setup required Other equipment
an aperture of f/16 in order to keep as much of the The coin was placed on an automatic turntable
subject in focus as possible. The Sony setup used that could be controlled via an app on a
an aperture of f/11. smartphone. There are several advantages
to using an automatic turntable, for example
Lighting manual intervention to turn the object after each
In order to provide consistent ambient light exposure is not required, and the number of stops
to minimise shadows being cast, an object the turntable makes can be controlled to give,
photography tent was used (Figure CS7.1). say, 12, 36 or 72 imaging positions for each full
rotation. For this object, 36 images per revolution
was selected.

< < Contents 107

 edges of the coin, best represented in the lowest
ring of imagery, were shot with the 200mm macro
lens. Very small, but clearly visible, points on the
edges were later used to match the two halves of
the coin together. Many of these points were not
visible to the naked eye but, after taking some test
images, it became clear that enough of them were
discernible for matching. Images were shot in RAW
on both cameras.

Processing
White balance was applied during RAW processing
to uncompressed TIFF format. The images
were then renamed according to local archival
convention and metadata added. No other pre­
processing was applied.

Photogrammetric processing was carried out

using Agisoft Photoscan Professional. Images
were examined and out-of-focus areas masked.
The images were aligned, and a medium-quality
dense cloud and mesh produced. This allowed for
the semi-automated identification of the targets
on the scales. After adjusting their positions
in the images as necessary, the control points
Figure CS7.2 (top) were given coordinate values and the alignment
Scales and grey card next to the object during optimised. The model for each face of the coin
photography.
was then scaled correctly. The reconstruction
Figure CS7.3 (bottom) region was adjusted inwards to include only the
The imaging strategy used for each half. coin. The dense cloud and mesh were recomputed
using high-quality settings. The resulting meshes
had face counts of circa 2 million polygons after
trimming out any extraneous geometry such as
Other objects were placed adjacent to the coin the surface of the turntable.
(Figure CS7.2), and a forensic scale, with a
manufacturer’s quoted accuracy of 0.1mm, was The chunks representing the two halves were then
used to provide scale and orientation for each duplicated to avoid having to reprocess them, and
model. One shot was taken with a full-colour subsequent work took place on the copies. The
reference card and a grey card in view. The grey intention was to match the two halves of the coin
card was left in for all subsequent photography using a marker-based approach, ie identifying
because removing it was not necessary. common points on both models, which were
only around the edges, and using them to align
Imaging strategy the two models correctly. Existing control points
Figure CS7.3 illustrates the imaging strategy. The for the two halves were removed, as they would
obverse and reverse faces of the coin were imaged have had an adverse effect on correct alignment.
separately. Three rings of photography were taken Markers were added (Figure CS7.4), again in a
for each side, with 36 images in each ring, and a semi-automated process, which required only the
vertical shot was taken with the camera directly subsequent removal of markers where they were
over the coin, offset from the tripod on a bar. The not visible. The two halves were then aligned

< < Contents 108

using the marker-based method while keeping
their scale fixed, correctly placing them relative
to each other.

The two models were exported as .ply format files

to Geomagic Wrap. This was necessary because
the models were intended for 3D printing, for
which the meshes would require some additional
editing to make them suitable. It was also so that
the polygons in the meshes representing the two
halves could be stitched together to produce a
single seamless model with no holes or artefacts
likely to cause problems in the 3D printing
process, such as self-intersections in the mesh.

In Agisoft Photoscan, the models of the two

halves, now aligned, were merged to form a new
model of both halves. The mesh was deleted
and replaced with the edited model from
Geomagic, which was on the same coordinate
system and at the same scale (Figure CS7.5
shows the final model).

Products
The wider project (the coin was one of several
objects being modelled) has yet to be finished
(May 2017). The coin is to be 3D printed at an
enlarged scale, and used for online presentation Figure CS7.4 (top)
via the Sketchfab website. The two dissemination Markers used to match the two halves of the model.
methods have different requirements in terms Figure CS7.5 (bottom)
of polygon count. Higher polygon count models The final complete model.
are necessary for the 3D printing, especially as
it involves enlarging the object, while relatively
low polygon count models are suitable for online
viewing. The combined mesh model was therefore
duplicated for decimation to a number of different
polygon target counts for the different outputs,
and each model textured before export.

< < Contents 109

5 References

ACCORD project 2015 Available at: https:// Dunn, C and Pearson, T 2005 Rievaulx Abbey Helmsley
accordproject.wordpress.com/, accessed 16/03/2016 North Yorkshire Level 1 Field Survey Report 2005.
Swindon: English Heritage
Andrews, D, Bedford, J and Bryan, P 2015 Metric Survey
Specifications for Cultural Heritage, 3rd edn. Swindon: English Heritage 2001 Conservation Plan, Rievaulx
Historic England. Available at: https://historicengland. Abbey, North Yorkshire. Swindon: English Heritage
org.uk/images-books/publications/metric-survey­
specifications-cultural-heritage/ English Heritage 2009 Measured and Drawn: Techniques
and Practice for the Metric Survey of Historic Buildings,
Barnes, A Close-Range Photogrammetry: A Guide 2nd edn. Swindon: English Heritage. Available on
to Good Practice. Available at: http://guides. request from: [email protected]
archaeologydataservice.ac.uk/g2gp/Photogram_Toc,
accessed 22/02/2016 English Heritage 2010a The Light Fantastic: Using
Airborne lidar in Archaeological Survey. Swindon:
Bradski, G and Kaehler, A 2008 Learning OpenCV:
English Heritage. Available at: https://historicengland.
Computer Vision with the OpenCV Library. Sebastopol,
org.uk/images-books/publications/light-fantastic/
CA: O’Reilly

English Heritage 2010b Thornton Abbey, North

Brumana, R, Oreni, D, Van Hecke, L, Barazzetti,
Lincolnshire: Historical, Archaeological and Historical
L, Previtali, M, Roncoroni, F and Valente, R 2013
Investigations. Research Department Report
‘Combined geometric and thermal analysis from UAV
Series 100-2010. Swindon: English Heritage.
platforms for archaeological heritage documentation’.
http://research.historicengland.org.uk/Report.
ISPRS Ann Photogramm Remote Sens Spatial Inf Sci II­ aspx?i=14941&ru=%2fResults.aspx%3fp%3d1%26n%
5-W1, 49–54 3d10%26t%3dThornton%2520Abbey%2c%2520North
%2520Lincolnshire%26ns%3d1
Calarco, A, Dam, A and Stewart Walker, A 2004
‘Geometric precision of scanned imagery for English Heritage 2011 3D Laser Scanning for Heritage,
production photogrammetry’. ISPRS Archives XXXV 2nd edn. Swindon: English Heritage. https://
(B1), 59–62 www.historicengland.org.uk/images-books/
publications/3d-laser-scanning-heritage2/
Chandler, J H 2010 ‘Tips for the effective use of close
range digital photogrammetry for the Earth sciences’. Fergusson, P and Harrison, S 1999 Rievaulx Abbey:
ISPRS Commission V, Working Group V/6. Available Community, Architecture Memory. Yale: Yale University
at: http://isprsv6.lboro.ac.uk/tips.html, accessed Press
24/08/2014
Fischer, M A and Bolles, R C 1981 ‘Random sample
Dawson, T and Winterbottom, S 2003 Airborne Remote consensus: a paradigm for model fitting with
Sensing and Ground Penetrating Radar Survey. Coll and applications to image analysis and automated
Tiree: Historic Scotland cartography’. Commun ACM 24, 381–95

Doneus, M, Verhoeven, G, Fera, M, Briese, C, Kucera, M Fryer, J G, Mitchell, H L and Chandler, J H (eds)
and Neubauer, W 2011 ‘From deposit to point cloud: 2007 Applications of 3D Measurement from Images.
a study of low-cost computer vision approaches for Caithness: Whittles Publishing
the straightforward documentation of archaeological
excavations’, in Pavelka K (ed) Geoinformatics. Prague:
Faculty of Civil Engineering, Czech Technical University,
6, 81–8

< < Contents 110

Garmin 2007 GPSMAP 496 Portable Aviation Receiver McCarthy, J 2014 ‘Multi-image photogrammetry
Owner’s Manual. Available at: http://static.garmin. as a practical tool for cultural heritage survey and
com/pumac/GPSMAP496_OwnersManual_forEurope_. community engagement’. J Archaeol Sci 43, 175–85
pdf, accessed 12/05/2016
McCarthy, J and Benjamin, J 2014 ‘Multi-image
Gray, S UAV survey: A Guide to Good Practice. Available photogrammetry for underwater archaeological site
at: http://guides.archaeologydataservice.ac.uk/g2gp/ recording: an accessible, diver-based approach’. J
AerialPht_UAV, accessed 22/02/2016 Maritime Arch 9 (1), 95–114

Green, S, Bevan, A and Shapland, M 2014 ‘A McCarthy, J, Robertson, P and Mackay, E 2015
comparative assessment of structure from motion ‘Discovery and survey and of a 17th–18th century
methods for archaeological research.’ J Archaeol Sci 46, shipwreck near Drumbeg, NW Scotland: an initial
173–181 report’. Int J Nautical Arch 44, 202–8

Grun, A, Remondino, F and Zhang, L 2004 Mitrovic, M, Cvijetinovic, Z and Mihajlovic, D 2004
‘Photogrammetric reconstruction of the Great Buddha ’Procedures and experiences on using desktop scanner
of Bamiyan, Afghanistan.’ The Photogrammetric Record for orthophoto production. ISPRS Archives XXXV (B1),
19 (107), 177–99 53–8

Hartley, R and Zisserman, A 2003 Multiple View NADRAP Available at: http://archaeologydataservice.
Geometry in Computer Vision. Cambridge: Cambridge ac.uk/era/section/record_manage/rm_projects_
University Press nadrap_home.jsf, accessed 15/01/2016

Hass, C 2007 JPEG Compression, Quality and File Size. Neumann, K 2009 ‘Digital aerial cameras.’ Proc 24th Int
Available at: http://www.impulseadventure.com/ Cartographic Conference. Santiago, Chile. Available at:
photo/jpeg-compression.html, accessed 27/08/15 https://www.semanticscholar.org/paper/Digital-Aerial­
Cameras-Neumann/365b2985ee512c56a2b0be3816efb
Historic England 2015 Where on Earth are We? The Role 4cac6f32849, accessed 01/12/2016
of Global Navigation Satellite Systems in Archaeological
Field Survey. Swindon: Historic England. https:// Nocerino, E, Menna, F, Remondino, F and Saleri, R 2013
historicengland.org.uk/images-books/publications/ ‘Accuracy and block deformation analysis in automatic
where-on-earth-gnss-archaeological-field-survey/ UAV and terrestrial photogrammetry: lesson learnt’.
ISPRS Annals II-5/W1 203–8
ICAO Circular 328 AN/190 Unmanned Aircraft Systems
(UAS). Available at: https://www.icao.int/Meetings/ Papworth, H, Ford, A, Welham, K and Thackray, D 2015
UAS/Pages/UAS_Documents.aspx, accessed ‘Assessing 3D metric data of digital surface models for
15/07/2016 extracting archaeological data from archive stereo-
aerial photographs.’ J Archaeol Sci 72, 85–104
Knott, P 2013 ‘Gun Rocks, Farne Islands (undesignated
site assessment)’. WA Ref 83803.12. Unpublished Powlesland, D, Lyall, J and Donoghue, D N M 1997
fieldwork report for Wessex Archaeology on behalf of ‘Enhancing the record through remote sensing.
English Heritage. The application and integration of multi-sensor,
non-invasive remote sensing techniques for the
Lowe, R G 2004 ‘Distinctive image features from scale- enhancement of the Sites and Monuments Record.
invariant keypoints’. Int J Comput Vis 60, 91–110 Heslerton Parish Project, N. Yorkshire, England.’
Internet Archaeology 2. Available at: http://intarch.
Luhmann, T, Robson, S, Kyle, S and Harley, I 2006 ac.uk/journal/issue2/index.html
Close Range Photogrammetry: Principles, Methods and
Applications. Caithness: Whittles Publishing Remondino F, Del Pizzo S, Kersten P T and Troisi S 2012
‘Low-cost and open-source solutions for automated
Mallinson H and Wings, O 2014 ‘Photogrammetry in image orientation: a critical overview’. Lect Notes
paleontology: a practical guide’. J Paleontol Tech 12, Comput Sci 7616, 40–54
1–31
RICS 2010 Vertical Aerial Photography and Digital
McCarthy, J 2012 ‘Unknown Wreck, Drumbeg Imagery, 5th edn. London: RICS
(undesignated site assessment)’. WA Ref 83800.15.04.
Unpublished fieldwork report for Wessex Archaeology Stylianidis E and Remondino F (eds) 2016 3D Recording,
on behalf of Historic Scotland. Documentation and Management of Cultural Heritage.
Caithness: Whittles Publishing

< < Contents 111

TheoLt 2010 Photography For Photoplan3d: The 3×3 Waldhäusl, P and Ogleby, C 1994 ‘3x3-rules for simple
Rules. Available at: http://www.theolt.com/web/ photogrammetric documentation of architecture’.
photography-for-photoplan3d-the-3x3-rules/, ISPRS Archives XXX, 426–9
accessed 22/12/16
Wu, C, Agarwal, S, Curless, B and Seitz, S 2011
TSA 2013 Client Guide to Small Unmanned Aircraft ‘Multicore bundle adjustment.’ IEEE Proc CVPR 2011,
Surveys, Issue 2. Newark on Trent: TSA. Available at: 3057-3064
http://www.tsa-uk.org.uk/for-clients/guidance­
notes/, accessed 01/12/2015

TSA 2015 Client Guide to Digital Ortho-Photography,

Issue 3. Newark on Trent: TSA. Available at: http://www.
tsa-uk.org.uk/for-clients/guidance-notes/, accessed
01/12/2015

Tucker, C and Garratt, M 1977 ‘Leafoptical system

modeled as a stochastic process.’ Appl Optics 16, 635.

Verhoeven, G J 2007 ‘Becoming a NIR-sensitive aerial

archaeologist’. Proc SPIE 6742

Verhoeven, G J 2008 ‘Imaging the invisible using

modified digital still cameras for straightforward and
low cost archaeological near-infrared photography’. J
Archaeol Sci 35, 3087–100

Verhoeven, G J 2010 ‘It’s all about the format:

unleashing the power of RAW aerial photography’. Int J
Remote Sense 31, 2009–42

Verhoeven, G J 2012 ‘Near-infrared aerial crop mark

archaeology: from its historical use to current digital
implementations’. J Archaeol Method Theory 19, 132–60

Verhoeven, G J 2016 Chapter 4, in Stylianidis E and

Remondino F (eds) 3D Recording, Documentation and
Management of Cultural Heritage. Caithness: Whittles
Publishing

Verhoeven, G, Sevara, C, Karel, W, Ressl, C, Doneus,

M and Briese, C 2013 ‘Undistorting the past: new
techniques for orthorectification of archaeological
aerial frame imagery’, in Corsi, C et al (eds) Good
Practice in Archaeological Diagnostics, Natural Science
in Archaeology. Switzerland: Springer International
Publishing

Wackrow, R 2008 ‘Spatial measurement with consumer

grade digital cameras’. Unpublished PhD thesis,
Loughborough University. Available at: https://dspace.
lboro.ac.uk/dspace-jspui/handle/2134/3827, accessed
10/12/2014

Wackrow, R, Chandler, J H and Gardner, T 2008

‘Minimising systematic errors in DEMS caused by an
inaccurate lens model’. ISPRS Archives XXXVII, 1–6

< < Contents 112

Software

Software mentioned in the text; this is not an endorsement.

Name By Found at

3DF Zephyr http://www.3dflow.net/3df-zephyr-pro-3d-models-from-photos/

3DHOP http://3dhop.net/
Aerial John Haigh [email protected]
Airphoto Irwin Scollar http://www.uni-koeln.de/~al001/airdown.html
Arc3D KU Leuven http://www.arc3d.be
ArcGIS ESRI http://www.esri.com/arcgis/about-arcgis
Bundler https://github.com/snavely/bundler_sfm
Cloud Compare http://www.danielgm.net/cc/
ContextCapture Bentley https://www.bentley.com/en/products/product-line/reality­
modeling-software/contextcapture

Correlator 3D SimActive http://www.simactive.com/en/

Cyclone Leica http://leica-geosystems.com/en-gb
Drone Deploy www.dronedeploy.com
Erdas Imagine Hexagon http://www.hexagongeospatial.com/products/power-portfolio/
erdas-imagine
GRASS https://grass.osgeo.org/
Lidar Visualisation http://www.arcland.eu/outreach/software-tools/1806-lidar­
toolbox visualisation-toolbox-livt
Meshlab http://meshlab.sourceforge.net/
Netfabb https://www.autodesk.co.uk/products/netfabb/overview
Photomodeler http://www.photomodeler.com/index.html
Photoplan FARO http://www.faro.com/products/construction-bim-cim/faro­
photoplan/
Photoscan (Pro) Agisoft http://agisoft.com/
Pix4D mapper https://pix4d.com/
QGIS http://www.qgis.org/en/site/
QT modeller http://appliedimagery.com/
RealityCapture https://www.capturingreality.com/
Recap 360 Autodesk https://www.autodesk.com/products/recap/overview
Recap Image Autodesk https://www.autodesk.com/products/recap/overview
Relief Visualisation http://iaps.zrc-sazu.si/en/rvt#v
Toolbox (RVT)
Scene Faro http://www.faro.com/en-us/products/faro-software/scene/
overview
Sketchfab www.sketchfab.com
Visual SfM http://ccwu.me/vsfm/

< < Contents 113

Other web resources

Aerial Archaeology http://www.univie.ac.at/aarg/

Research Group (AARG)
Archaeology Data http://archaeologydataservice.ac.uk/era/section/record_manage/rm_projects_
Service (ADS) nadrap_home.jsf
Bing maps https://www.bing.com/maps
CAA www.caa.co.uk/uas
GSD calculator https://support.pix4d.com/hc/en-us/articles/202560249-TOOLS-GSD-Calculator
International Centre http://www.iccrom.org/
for the Study of the
Preservation and
Restoration of Cultural
Property (ICCROM)
International Society http://www.isprs.org/
for Photogrammetry
and Remote Sensing
(ISPRS)
Kite Aerial http://www.kiteaerialphotography.org.uk/
Photography
Royal Institution of https://www.rics.org/uk/
Chartered Surveyors
(RICS)
Remote Sensing and http://www.rspsoc.org.uk/
Photogrammetry
Society (RSPSoc)

< < Contents 114

6 Glossary

2.5D Breaklines
Notation used to describe points, or a surface made Lines used to define sharp changes of slope in a DEM.
up of points, that have plan coordinates and a height
value but are not part of a true 3D surface; there is no Bundle adjustment
possibility of undercuts. A process that adjusts the ‘bundles’ of rays between
each camera centre and a set of projected 3D points
360 degree camera until a minimal discrepancy between the positions of
A camera that can produce a cylindrical image with one the observed and re-projected points is achieved.
exposure either using multiple lenses or by using one
lens that automatically pans around 360 degrees. Bundle block adjustment
Bundle adjustment applied to a block of images, ie a
Absolute accuracy number of strips.
The accuracy with respect to a defined coordinate
system. BVLOS
Beyond visual line of sight. The distance beyond which
AF it would not be possible to see an SUA.
Auto-focus.
CAD
Affine transformation Computer-aided design.
A transformation that will fit any three points, in 3D
space, to any other three. Angles and distance between Calibrated focal length
the points will not be maintained but parallelism An accurately measured focal length.
between any two lines will be.
Camera shake
AGL see motion blur
Above ground level.
CFA
Aperture Colour filter array. An arrangement of colour filters on a
The, usually, adjustable opening through which light digital camera sensor that means certain image diodes
passes into a camera. only receive particular colours. The resulting colour
channels, eg RGB, are used to form
Bayer array the final image.
The particular arrangement of colour filters used in
most digital camera sensors; there are twice as many Channel
green filters as red or blue; see also CFA. The units of separation of a digital image, eg RGB.

Bits Chief ray

The basic unit of information in computing; A light ray that theoretically passes in a straight line
it can have only one of two values. from the object point through the perspective centre of
the lens and onto the image plane of a camera.
Bowl effect
The situation where a computed model or surface Chromatic aberration
bends up at the sides or ends as a result of The result of variation in the focusing of different
accumulated errors, so that it is not as flat colours on an image sensor leading to noticeable
as it should be. coloured fringes.

bpp
Bits per pixel; the higher the value, the more colour
variation is encoded in an image.

< < Contents 115

Clipped tonal curve Diffraction limit
The result of capturing an image where the intensity The point beyond which loss of sharpness because
of certain parts is outside the range that can be of diffraction becomes unacceptable when reducing
represented. This can lead to, for example, clipped aperture size.
highlights, where the full range of brightness in the
subject is not apparent in the resultant image. Dishing effect
see bowl effect
Collinearity
The situation in which a number of points lie on the dpi
same single straight line. Dots per inch. A definition of the resolution of an
image originally used by the printing industry where
Colour saturation the dots are dots of ink. In a digital image the dots are
The intensity of a particular colour in an image. analogous to pixels; the higher the dpi, the more detail
The primary colours, red, yellow and blue, are is represented in the image.
fully saturated.
Drone
Colour temperature see SUA
An expression of the colour of a light source based on
the colour of a theoretical black body when heated to a DSM
certain temperature (degrees Kelvin). A reddish-yellow Digital surface model. For landscapes, the DSM is the
white light is cooler than a bluish white light, while surface including features such as buildings and trees,
daylight has the highest colour temperature. while the DTM represents the ‘bare earth’ surface
resulting from the filtering out of these features.
Control
Points with known coordinates used to position or DTM
constrain a plan or 3D model. Digital terrain model. For landscapes, the DTM
represents the ‘bare earth’ surface resulting from the
Convergent filtering out of features such as buildings and trees,
In the case of photogrammetry, the situation where while the DSM is the surface including features such as
the camera axes are not parallel, ie they are pointing buildings and trees.
towards the same part of the subject.
Dynamic range
CT scanner The ratio between the maximum and minimum
Computed tomography scanner. An instrument that luminance in an image. In a high dynamic range image,
generates 3D images by combining numerous slices of a greater range of luminance will be correctly exposed.
an X-ray image.
E57
Decentring lens distortion parameters (p1, p2) A non-proprietary format for point cloud
see tangential distortion data, developed by the American Society for Testing
and Materials.
DEM
Digital elevation model. A digital representation of a EGNOS
surface. DSM and DTM are European Geostationary Navigation Overlay Service.
types of DEM. A form of differential GPS where the corrections are
transmitted from geostationary satellites.
Depth of field
That portion of the field of view of a camera that will Electronic front and rear curtain
be in focus for a particular aperture setting. A smaller A feature in some digital cameras that in effect
aperture gives a greater depth of field. speeds up the opening and closing of the shutter by
sequentially activating the image diodes just ahead of
DGPS the movement of the physical curtain.
Differential GPS. A way of improving the accuracy
of coordinates collected with a GNSS by using two EXIF
receivers, one of which is located at a known point. Exchangeable image file format. A standard
The relative accuracy between the two receivers will be for digital image metadata such as date and
high so that the data collected by the fixed receiver can exposure information.
be used to correct that captured by the moving one.
The fixed receiver is often replaced by a virtual signal
transmitted from a server.

< < Contents 116

Exposure GNSS
The act and associated settings of opening the shutter Global navigation satellite system. A system that
in a camera to expose the sensor to light; also used to enables surveying or navigation by reference to a
describe the result of opening the shutter, ie an image. number of satellite constellations.

f/number, f/stop GNU

In effect the size of the aperture used for a particular A free computer operating system.
exposure, but more correctly the ratio of the focal
length of a camera lens to the diameter of the aperture. GPR
Ground-penetrating radar. A geophysical process that
FAPAR uses radar pulses to image surfaces below ground.
Fraction of absorbed photosynthetically active
radiation. A method of quantifying the amount of solar GPS
radiation absorbed by leaves for photosynthesis. Global positioning system. A generic term used to
describe surveying or navigation by reference to a
Fast lens satellite constellation, although it is specifically the
A lens that has the ability to use a larger aperture name for the satellite constellation operated by the
(eg f/1.2), so, ie more light will enter the camera during USA; see also GNSS.
a shorter exposure time.
Ground-truthing
Fiducial marks The process of verifying remotely sensed data by
Marks that appear in the frame of an analogue checking (a sample of) the findings on the ground.
metric camera to enable correction of film
distortion and placement of the image in a GSD
photogrammetric instrument. Ground sample distance. The distance on the ground
or subject that is represented by the distance between
Firmware adjacent pixel centres in an image.
Software permanently programmed into a piece
of equipment. GUI
Graphical user interface. The means by which the user
Fish-eye lens communicates with most software.
An ultra-wide angle lens, usually giving a field of view
of 180 degrees or more. Hachures
Symbols used to indicate direction and steepness of
Flat lighting slope on a map or topographic survey.
An arrangement of lights that results in little or no
shadow on the subject. Hand-drawn survey
Survey undertaken using on-site drawing and hand
Focal length measurement, usually with a tape measure.
The distance between a lens and the image sensor
when the subject is in focus. HDR
High dynamic range; see dynamic range.
Fx and Fy
Focal length in pixels. Although a lens has only one Height displacement/distortion
focal length, F, if the pixels on camera sensor are Errors in an image caused by variation in the relief
rectangular then fx and fy are F multiplied by the of the subject. Those parts of the subject nearer the
number of pixels per unit length along each axis. camera will be at a larger scale than those further
away and their position will also be more adversely
GCPs affected by tilts of the camera.
Ground control points. Control points used in the
mapping of landscapes from aerial platforms. Hillshade
A method of depicting slope in a graphical
GIS representation of a landscape.
Geographic information system. A database where
the information is related to a map or other graphical Histogram equalisation, histogram stretching
representation of the surface of the Earth. A method for improving the contrast of a digital image.

< < Contents 117

Hot mirror ISO value
A filter normally used to protect optical systems from International Standards Organisation value. A standard
infra-red radiation. for describing image sensor sensitivity to light,
analogous to film speed. The higher the ISO value, the
Hot shoe less light is required for the same exposure.
A connector, usually on the top of a camera, to allow
the mounting of ancillary equipment such as a flash Key points
gun. The term ‘hot’ refers to the fact that the mount see IPs
enables the firing of the flash in synchronisation with
the shutter release. KML file
A version of XML developed for viewing geographical
Image diodes data in Google Earth.
The individual receptors in an image sensor; analogous
to the pixels in the resultant image. Landsat
A satellite system developed by the National
Image distance Aeronautics and Space Administration (NASA) to
The distance between a point on the subject of a provide imagery of the surface of the Earth for resource
photograph and the image plane. management purposes.

Image noise Laser scanner

Unwanted variations in an image resulting from the A laser device that collects 3D coordinates of a given
signal noise in the sensor. region of a surface automatically and in a systematic
pattern at a high rate (thousands of points per second),
Image pair achieving the results in (near) real time.
see stereo pair
Lens distortion
Image plane Distortion of an image caused by characteristics of
That part of a camera where the image is projected and the lens.
recorded, either onto film or a sensor.
Lidar
Image point Light detection and ranging. A system that uses laser
The point on an image representing a specific point on pulses to measure the distance to an object or surface,
the subject. typically determining the distance by measuring
the time delay between transmission of a pulse and
Image rectification detection of the reflected signal.
see rectified photography
Luminance
Image triangulation The intensity of light emitted from a surface per
A technique used to calculate the relative position for unit area.
each image in every pair in a strip of images.
Macro lens
Inner or interior orientation The process of A lens that results in an image with a scale of 1:1
determining the internal characteristics of a camera or larger.
system, such as the focal length and lens distortion.
Mesh
INS A method of digitally representing a surface using
Inertial navigation system. A computerised system points connected by lines to define a large number of
using accelerometers and gyroscopes to calculate, smaller polygons (usually triangles or squares).
by dead reckoning, the path of a vehicle on which it
is mounted. Metadata
Data about data, eg exposure information for a
IPs digital image.
Interest points. Points on an image identified by
photogrammetric software for the image-matching Metric camera
process. A camera with a calibrated lens.

IS
Image stabilisation. A system inside a camera that
moves the sensor while the shutter is open in an
attempt to reduce motion blur.

< < Contents 118

Mirror lock-up mode Perspective distortion
The situation where the mirror in a single lens reflex Distortion of the subject of an image caused by the
camera is moved and locked out of the image path use of a perspective projection, as opposed to an
before the shutter is opened. This removes a source of orthogonal projection.
possible camera shake.
Photo-diodes
Mirrorless camera see image diodes.
A camera where there is no mirror, to allow a through-
the-lens viewfinder. The viewfinder is either digital or Photogrammetric reconstruction
uses separate optics. The process of producing a model using
photogrammetry.
Motion blur
Blur in an image caused by movement of the camera Pixel
(camera shake) during exposure. The smallest element of a digital image, analogous to
an image diode in a digital camera sensor.
MP
Mega-pixel. Point cloud
A set of, usually many, points in a 3D coordinate system
MVS used to represent the surface of a subject.
Multi-view stereo. A photogrammetric process using
multiple convergent images. Post spacing
The distance between points in a DEM or point cloud.
NDVI
Normalised difference vegetation index. A method for Posterisation
measuring vegetation health where a near infra-red The reduction in tonal range of an image resulting in
channel is used in addition to the red channel in a RGB sharp changes of colour. The term refers to a deliberate
image. In essence, NDVI = (NIR-red)/(NIR +red). act to facilitate the printing of posters.

NIR ppm
Near infra-red. Parts per million.

NIR pass filter Prime lens

A filter that allows NIR radiation to pass through it. A fixed focus lens.

Object point Principal distance

The point on the subject of a photographic image The equivalent of the focal length of a camera.
represented by the image point on the sensor.
Principal point
Open GL The point where a straight line passing through a lens
The industry-standard graphics programing language. at right angles to the imaging plane meets that plane.

Orthogonal projection Principal point offset

A method of representing a 3D subject where all the The extent to which the principal point is not in the
projection lines are at right angles (orthogonal) to the centre of the image plane.
projection plane.
Projection centre
Ortho-images The theoretical centre of a camera lens system through
Images resulting from an orthogonal projection. which all light rays from the subject pass before
arriving at the sensor.
Perspective centre
The point of origin or termination of bundles of rays Quantisation
or projecting lines directed to a point object in a A matrix that controls the compression ratio of, for
camera system, effectively the projection centre of an example, a JPG image.
ideal camera.
Radial lens distortion parameters (k1, k2, k3, k4)
Perspective correction lens Distortions that vary with distance from the centre of
see tilt-shift lens the lens.

< < Contents 119

RANSAC Ring flash unit
RANdom SAmple Consensus. An algorithm for A ring-shaped photographic flash that mounts around
detecting outliers in a set of points. the lens, thus giving a more uniform illumination of
the subject.
Raster image
A digital image composed of individual pixels; a Rising front lens
straight line will be represented by several pixels, see tilt-shift lens.
whereas in a vector plan it will consist of two
connected points. RMSE
Root mean square error. A statistical method of
RAW format measuring the difference between the measured and
Various proprietary image formats (not an acronym) predicted values of a sample; often used in surveying
containing minimally processed data from an image as an indication of accuracy.
sensor. The files need to be post-processed to produce
images in common formats, which means exposure, RPAS
colour adjustments, etc, can be made. Each camera Remotely piloted aircraft system; see SUA.
manufacturer has its own type of RAW file.
Sensor
Rectification plane An electronic component or system for detecting
The plane that a rectified image is corrected to fit. Any particular values in its environment. In the case of
features in the image not on that plane will not be to cameras, the values will be for light.
the specified scale.
Sensor sensitivity
Rectified photography, rectified image The change in output of the sensor per unit change in
The process and image resulting from correcting a the parameter being measured, eg for a digital camera
single image to fit a 2D plane. the variation in colour required to give a different pixel
value in the resulting digital image.
Reference plane
see rectification plane. Shutter speed
A way to describe the length of time a camera shutter
Reflectorless is open.
The ability of an electromagnetic distance
measurement system, in, for example, a TST, to Side lap
measure to any surface rather than just to a prism. The overlap between strips of photography.

Relative accuracy SIFT

The accuracy of one element of a survey with regard to Scale invariant feature transform.

another rather than a particular coordinate system. An algorithm for detecting similar features in a number

of images; used in photogrammetric applications for

Relief displacement image matching.

see height displacement.

Signal noise
Re-projection error Unwanted and usually unknown variations that a signal
The geometric error as a result of the difference in may suffer in the collection process. This can occur
image distance between a projected point and a during capture, transmission or processing.
measured point.
Specular
Resolution Mirror-like.
The smallest interval measurable by an instrument
such as a scanner or camera. Radiometric resolution Stand-off
refers to the number of different colours that can be The distance away from a subject that it is possible or
captured, while geometric resolution refers to the desirable to take a measurement or image from.
physical size of the smallest measurable element.
Stereo pair
RGB Two photographs with sufficient overlap for
Red, green, blue. The usual colour channels of a photogrammetric application.
digital image.

< < Contents 120

Stereo photogrammetry Topographical relief
Photogrammetry that employs two rather than multiple Variations in the shape of a surface; usually height in
images to calculate the position of the subject. a landscape.

Stopping down Trilateration

The act of reducing the camera aperture; increasing The determination of the position of a point by
the f/number. measuring three or more distances to other
known points.
SUA
Small unmanned aircraft. Small aircraft weighing TST
less than 20kg that are controlled remotely from the Total station theodolite. A surveying instrument
ground or with a pre-programmed flight path. used to measure angles (vertical and horizontal) and
They can be rotary, like a helicopter, or fixed-wing, distances and record them. Such instruments are
like an aeroplane, and usually carry a camera or some known as total stations because all the data required is
other instrument. collected and stored in one self-contained system.

Swath UAV
In the context of aerial photography or lidar, the strip Unmanned aerial vehicle; see SUA.
of ground covered by one pass of the aircraft.
Unsharp masking
Systematic error A process available in some image-processing
Errors that are constant or a constant ratio, rather than softwares for sharpening images. It refers to the use of
random, and hence can be corrected for. a blurred negative of the image to create the mask.

Tangential distortion Vector plan

Image distortion resulting from the camera lens and A drawing or map made up of vectors, ie lines
image plane not being parallel. connecting points, rather than a raster image, where a
straight line can be represented by numerous pixels.
Textured mesh
A mesh with an associated image that is mapped to Vectorisation
the surface to provide a more realistic rendering of The act of converting a raster image into a series
the subject. of vectors.

Tie points Viewshed

Points with unknown coordinates that can be used The part or parts a of landscape that are visible from
to tie together two elements of a survey, such as a particular vantage point.
photogrammetric models.
VR
TIFF Vibration reduction; see image stabilisation.
Tagged image file format. A non-proprietary digital
image format. WGS84
World Geodetic System 1984. The coordinate system
Tilt-shift lens used by GNSSs that is often transformed to a particular
A lens that can be tilted or shifted with respect to the national grid system for subsequent use.
image plane. Such lenses are generally used to correct
the perspective effect in architectural photography. White balance
The process of correcting colour balance in a digital
TIN image so that, for example, whites in the subject are
Triangulated irregular network. A type of mesh made actually white in the image.
up of points joined by lines of varying length, resulting
in triangles of various sizes and shapes. Wide-angle lens
A lens with a shorter than normal focal length (35mm
Tonal curve or less) that thus captures a wider field of view.
A curve on a graph representing the variations in tone
in an image (from dark to light) of shadow, mid-tones
and highlights.

< < Contents 121

Acknowledgements

David Andrews (Historic England), Paul Bryan

(Historic England), Pete Horne (Historic England),
John McCarthy (Wessex Archaeology), Andrew
Blogg (Future Aerial Innovations), Clive Boardman
(Imetria) and Eva Fairnell are thanked for their
comments and suggestions on the text. Heather
Papworth (Bournemouth University; case study
1), David Went (Historic England; case study 2),
Nick Hannon (case study 3), Dominic Powlesland
(Landscape Research Centre; case study 5) and
John McCarthy (Wessex Archaeology; case study
6) are thanked for providing material for the
case studies.

< < Contents 122

Contact Historic England

East Midlands South East

2nd Floor, Windsor House Eastgate Court
Cliftonville 195-205 High Street
Northampton NN1 5BE Guildford GU1 3EH
Tel: 01604 735460 Tel: 01483 252020
Email: [email protected] Email: [email protected]

East of England South West

Brooklands 29 Queen Square
24 Brooklands Avenue Bristol BS1 4ND
Cambridge CB2 8BU Tel: 0117 975 1308
Tel: 01223 582749 Email: [email protected]
Email: [email protected]

Fort Cumberland Swindon

Fort Cumberland Road The Engine House
Eastney Fire Fly Avenue
Portsmouth PO4 9LD Swindon SN2 2EH
Tel: 023 9285 6704 Tel: 01793 445050
Email: [email protected] Email: [email protected]

London West Midlands

Fourth Floor The Axis
Cannon Bridge House 10 Holliday Street
25 Dowgate Hill Birmingham B1 1TG
London EC4R 2YA Tel: 0121 625 6870
Tel: 020 7973 3700 Email: [email protected]
Email: [email protected]
Yorkshire
North East 37 Tanner Row
Bessie Surtees House York YO1 6WP
41-44 Sandhill Tel: 01904 601948
Newcastle Upon Tyne NE1 3JF Email: [email protected]
Tel: 0191 269 1255
Email: [email protected]

North West
3rd Floor, Canada House
3 Chepstow Street
Manchester M1 5FW
Tel: 0161 242 1416
Email: [email protected]

< < Contents 123

 This page is left blank intentionally

< < Contents 124

We are the public body that looks after

England’s historic environment. We champion
historic places, helping people understand,
value and care for them.

Please contact
[email protected]
with any questions about this document.

HistoricEngland.org.uk

If you would like this document in a different

format, please contact our customer services
department on:

Tel: 0370 333 0607

Email: [email protected]

All information and weblinks accurate at the

time of publication.

Please consider the environment before printing

this document

HEAG066
Publication date: v1.0 October 2017
© Historic England
Design: Historic England

< < Contents