See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/263368196

A feature-based anthropometry for garment industry

Article in International Journal of Clothing Science and Technology · February 2013

DOI: 10.1108/09556221311292183

CITATIONS READS
13 1,981

3 authors, including:

Jing-Jing Fang Ming June Tsai

National Cheng Kung University National Cheng Kung University
68 PUBLICATIONS 746 CITATIONS 42 PUBLICATIONS 1,288 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Jing-Jing Fang on 14 October 2014.

The user has requested enhancement of the downloaded file.

 The current issue and full text archive of this journal is available at
www.emeraldinsight.com/0955-6222.htm

IJCST
25,1 A feature-based anthropometry
for garment industry
Iat-Fai Leong, Jing-Jing Fang and Ming-June Tsai
6 Department of Mechanical Engineering,
National Cheng Kung University, Tainan, Taiwan
Received 12 October 2011
Accepted 4 June 2012 Abstract
Purpose – The purpose of this paper is to propose a method to automatically generate individualized
body size measurements from cloud point of a body scanner. It aims to propose a fast, reliable, and
unambiguous method to obtain human body measurements for use in the garment industry.
Design/methodology/approach – Based on a previous study by the authors, geometric features on
the scanned body are identified by computerized algorithms through mathematical definitions.
Feature lines situated on the human body surface are created as polylines that pass through the body’s
features and three types of computer measurements (tape-measurement, contour-measurement, and
linear-measurement) are provided.
Findings – By dividing the body surface into rectangular patches using the feature lines as
boundaries, the body can be reconstructed easily with a minimal amount of triangles while retaining
the essential shape. The proposed measuring method applies to most manual measurements used in
the garment industry. The authors evaluated the anthropometry variations of the same subject to
explore the reliability of the proposed method. It was found that the precision of the method is well
below the standard requirement of the traditional manual method.
Research limitations/implications – In this research, subjects were scanned in standing pose;
this pose minimizes regions obstructed by body parts and permits maximal acquisition of as many key
landmarks. Since the features are identified by geometric analysis without the need for marker
attachment, measurements of the required sitting position are impossible to obtain in the current
study.
Originality/value – Resolution of meshing can be changed according to application requirements.
Contrary to the traditional manual method, efficiency and precision are the advantages of the present
method.
Keywords Modelling, Garment industry, Body regions, Image processing equipment, Measurement,
Programming and algorithm theory, Feature-based, Anthropometry, Multi-resolution model
Paper type Research paper

1. Introduction
Anthropometric data is widely used in the field of human factor study for productions
design. Generally, body dimensions measurements are performed manually by
experienced persons. Measurement operations and results are subject to personal
judgment, and therefore, the quality of the data is inconsistent. For the last decade, 3D
body scanning has become a promising tool in computerized anthropometry, such as
Cyberware (2005), TC2 (2005), Inspeck (2005), TriForm (2005), Vitus (2005) and
International Journal of Clothing Tsai et al. (2003). However, converting the scanned data into useful information for
Science and Technology engineering usage is still difficult. The problem occurs because the raw data often
Vol. 25 No. 1, 2013
pp. 6-23 contains insufficient information. That data should be processed before being used in
q Emerald Group Publishing Limited
0955-6222
human factor design. For example, the data needs to be segmented into significant
DOI 10.1108/09556221311292183 parts, searched for the landmarks or features, and the dimensions measured.
 In order to efficiently analyze the body shape, the first step of utilizing the cloud point is A feature-based
to segment the data set into several subsets according to human anatomy. We employed a anthropometry
method to separate the data set into the torso, arms and legs by finding cusps on a slicing
plane (Nurre, 1997; Ju et al., 2000; Lin, 2003). Robinette et al. (1999) used a 3D body scanner
to collect body dimensions in the CAESAR project. By using a neural net to recognize the
markers attached on the surface of the body, landmarks can be located in three different
poses (Robinette et al., 2001). The feature extraction operation is automated, but markers 7
are positioned manually. Placement of the markers directly affects the practical positions
of the landmarks. In our previous study, we developed a marker-less extraction method
for the standing pose (Leong et al., 2007). The scanned data is first segmented into the
torso, arms, and legs. Then, each part of the segmentation is transformed into a cylindrical
range image. Finally, image processing techniques and geometric calculations are applied
to extract major landmarks and features on the 3D body surface.
For many applications, a surface representation of the human body is needed. Many
studies have been conducted to convert point clouds into polygonal meshes (Jones et al.,
1995; Wang et al., 2003; Dekker et al., 1999). The level of detail of these models can be
adjusted, but most of these meshes do not contain high level information about the size
of the human body. In order to preserve more useful information in the polygonal
meshes, we construct the meshed model in a number of rectangular patches. Each
patch is bounded by a set of respective feature lines that are extracted from the cloud
point. The sampling points we applied on the structure regions are equally spaced
along the body surface. The resulting mesh provides a better representation of the
human body.
One of the key applications of the whole body scan is to extract measurements
directly from the point cloud. The common practice is to identify data points that
define the features. These measurements can be obtained by summing the distances
between every neighboring point (Pargas et al., 1997; Robinette and Daanen, 2006;
Zhong and Xu, 2006). The results from this method may be degraded by two factors:
sample number and noise. If the sample number is too low, then the original properties
of the feature are lost. Noise in the cloud point may form a zigzagged line rather than a
smooth curve. To overcome these problems, the measurement method we applied
employs a B-spline reconstruction technique. Then, the lengths of these features’
measurements are obtained by numerical integration.
The results of this study is beneficial to the 3D garment design (Carrere et al., 2000),
sizing system used by the fashion industry (Ashdown and Loker, 2001) and ergonomic
facilities.

2. Feature-based structure
The whole-body scanner has been widespread in industries for many years. But human
body scans provide little other than a huge number of surface cloud points. In order to
provide useful information of the model, anthropometric landmarks need to be
extracted from the point clouds. Unfortunately, there is no unanimity of the definitions
or standards of body landmarks in the literature. In this paper, we define landmarks as
geometric vertexes on the human body and features as curves that pass through a few
landmarks and drawn on the body surface. Our definitions of features are based on the
descriptions in ISO (1989) and ASTM (1999). We followed the landmark identification
techniques (Leong et al., 2007) to translate the literature definitions into a set of
IJCST mathematical definitions. The 41 feature lines are the primary features on human body.
25,1 Most measurements on the human body needed by the industries are either a section of
the primary features or a combination of specific features segments. For instance, the
outside leg length is the tape-measurement between the side of the waist and the sole;
therefore, we combine the side seam line under the waist girth and the outside leg line
to form a feature for that measurement.
8 Although primary features are sufficient for general measurements and
anthropometry purposes, a polygonal mesh representation of the body can provide a
lot more information, such as the 3D geometry of a feature or a surface and the surface
topology for a certain region of the body. In order to retrieve anatomical features in
such meshing, we proposed a feature-based mesh structure. The mesh structure has
geodetic coordinates similar to the longitudinal and latitudinal lines of the Earth. All
feature lines are included in their respective longitudinal or latitudinal positions within
the meshes. The structure of the mesh is rather compact yet also includes all body
features and landmarks. Every geodetic coordinate has its own geometrical meaning
that represents a specific body portion. It is possible to extract the mesh portions of
interest for one’s own design purposes. The number of triangles in the mesh structure
may vary to suit different applications. For each mesh resolution, we set the number of
interior nodes according to average proportion of the human body.
In the following sections, we will describe how to generate the mesh for the torso
and limbs. Lastly, we will describe how to develop a mesh model with the desired
resolution.

2.1 Torso structure

The torso is the center of our body structure; most of the body’s landmarks situate in
this part of the body. The arms, legs, and head connect to the torso. Thus, the desired
mesh structure not only has to incorporate the landmarks, but also must preserve the
topological structure of the human body. In previous research, we located all the
landmarks defined by ASTM. The feature lines of the body can be categorized into two
types: 11 near horizontal lines and ten near vertical lines. The 11 near horizontal lines
include the crotch girth, hip girth, mid-waist girth, waist girth, underbust girth, bust
girth, armpit girth, shoulder lines, neck base line, and upper neck line. The ten near
vertical lines include the front center line, front princess lines, side lines, armscye
girths, back princess lines, and back center line. By drawing the features on the range
image as shown in Figure 1, the deformed cylinder is developed into a set of
quadrilateral patches. Each patch is enclosed by four feature line segments. Hence,
interior node distributions can be used to represent a quadrilateral pattern which will
be described in Section 2.3 in greater detail.

2.2 Limb structure

The arms and legs represent half of the features of the human body. Both the arm and
leg are composed of two deformed cylindrical sections. In this research, subjects are
asked to take an easy and relaxed standing position during the body scan. As shown in
Figure 2, the legs are slightly separated, the knee must not bend, and the arms are
curve slightly inward. We can regard the leg as a single cylinder and the arm as two
cylinders. In order to form quadrilateral patches on the extremities, we identify five
horizontal girths and one vertical line on each leg and three horizontal girths and one
 A feature-based
anthropometry

Side neck point Upper neck line 9

Lower neck line
Mid-shoulder point
Shoulder line

Front neck point

Armhole
Shoulder point
Armpit girth
Armpits Armpits Bust girth
Bust points
Underbust girth
Shoulder blade point

Waist girth

Mid-waist girth

Hip girth

Crotch point
Crotch girth
Figure 1.
Back Back Sideline Front Front Front Sideline Back Torso sections on a
centerline princess princess centerline princess princess range image
line lines lines line

vertical line on each arm. Defined by ASTM D5219-99, there are thigh girths,
mid-thigh girths, knee girths, calf girths, ankle girths, and outside leg lines on legs.
On the arms, there are upperarm girths, elbow girths, wrist girths, and outside arm
lines. By combining the latitudinal and the longitudinal features lines shown in
Figure 2, the cylinder-like limbs are segmented into a set of quadrilateral sections.
The same principle of the node laying technique is applied to the arm and leg sections,
and the structures for arms and legs are formed.

2.3 Multi-resolution mesh

In order to build a visible structure that sufficiently reflects both the body features and
represents the body surface, multi-resolution meshes are generated. The structure
mesh is constructed based on the following two principles:
(1) Feature-based. On the body surface, feature lines should pass through a few
specific body landmarks. They section the whole body into several regions as
IJCST
25,1
Armhole

Upper arm girth

10
Elbow girth

Outside arm
line

Wrist girth

Thigh girth

Mid-thigh girth

Knee girth

Calf girth

Outside leg line

Figure 2.
The feature lines on Ankle girth
human body

shown in Figure 2. Each region is enclosed by four segmented feature lines,

namely left, right, upper and lower boundaries.
(2) Equal chord length distribution. The sampling points we applied on the
structure regions are equally spaced in three dimensions.

Here, we give an example describing how to distribute the given number of meshes to
form a torso structure. Suppose n £ m denotes ranking of the mesh array. For meshing
process, the patch is initially composed of nothing but the feature lines as boundaries.
A spring-mass system is used to achieve equal-spaced distribution to obtain the desired
meshes. First, five new interior nodes are added to segment the region into four
quadrangles (Figure 3(a)). The initial positions of the interior nodes are randomly placed
on each of the four boundaries and one inside the patch. The nodes form a virtual
spring-mass system with the nodes as particles and the lengths in between as virtual
springs. Then, the spring energy minimization is applied to the virtual spring-mass
system. Each sub-patch is further refined until the desired resolution is met (Figure 3(b)).
As shown in Figure 3, the corner vertices are the intersections of feature lines or the
resultant nodes of the previous iteration; thus they are regarded as stationary nodes.
 A feature-based
anthropometry

11

(a) (b)
Notes: (a) Initial node distribution; (b) energy minimization applying in solid lines Figure 3.
The spring-mass system
and then its recursive iteration in dash lines, and so on

In the energy release process, the edge nodes are moved along the edge of the region,
and the center node is moved inside the region. The following equations are applied to
obtain the spring energy in a single mesh. The spring energy of node Pi is:

X
m
E k ðP i Þ ¼ ð1=2KðjP i P j j 2 dj Þ2 Þ ð1Þ
j¼1

and the spring force acting on the node is:

X
m
f ðP i Þ ¼ ðKðjP i P j j 2 d j Þn~ P i P j Þ ð2Þ
j¼1

where K is the spring constant, jPiPjj denotes the length between Pi and Pj along the
body surface, dj is the!average length between every two neighboring points, n~ P i P j is
the unit vector of P i P j , m is the number of vertex that connected to Pi. In the
quadrilateral grid system, m is 8 for the center node and 3 for the boundary nodes.
Forces are generated by the springs connected to the node. The forces acting on the
center node pull the node along the body surface while the edge nodes are constrained
on the boundaries. After the elastic deformation energy reduces to a minimum, we
obtain an equal chord length nodes distribution map. The Lagrange equation is used
on the spring-mass system for energy releasing:


d ›L ›L
2 ¼Q ð3Þ
dt ›q_ i › qi

where qi is the generalized coordinate and the Lagrangian L is given by:

L¼T 2V ð4Þ
IJCST and Q is the external force applied on the node. In equation (4), T is the kinetic energy
of the motion of each node, and V is potential energy (spring energy). In the
25,1 spring-mass system, T, V and Q are given by:

1
T ¼ M ð_q 2 Þ
2
12 1
V ¼ KðjP i P j j 2 dj Þ2
2
d
Q ¼ D ðjP i P j j 2 d j Þ
dt
where M, D, K are spring mass, damping and stiffness coefficients, respectively. q
denotes the position of the node. In this case, no external force is applied on the system,
and the velocity of mass is assumed to be rather small so that the damping item is
negligible. Equation (3) can thus be simplified to:


d ›L ›L
2 ¼0 ð5Þ
dt ›qi_ › qi

For each region, the spring forces applied to each node are calculated. Then the nodes are
moved according to the amount of force applied. The energy release process continues
until the spring energy difference between two iterations is less than a given threshold.

2.4 Secondary features

In our previous study, the primary landmarks and feature lines are located on the body
scan data. We have incorporated those defined landmarks and feature lines in the
structure mesh. In reality, it is impossible to embed every required measurement on
these feature lines for all kinds of applications. To expand the applicability of the
structure mesh in further measurements, additional features can be extracted in an
efficient manner for their specific purposes. For instance, the across shoulder is defined
as a cross line that connects two acromions by passing through the cervical point
(Figure 4). This feature line is not located on the boundary of any patches, but key
landmarks are defined at the corners. Thus, we invoke a marching algorithm to draw
several piecewise linear curves on the structure mesh. By using the marching
algorithm, the user can define his required features on the meshes.

Cervical point

Figure 4.
The across shoulder Acromion
connects left and right
acromions by passing
through the cervical point
3. Anthropometry A feature-based
The purpose of body measurement is to extract the size information of a given feature. For anthropometry
different field of applications, there are various combinations of measurements are needed.
For instance, there are 41 pre-defined feature lengths on the body most commonly used by
the fashion industry. Therefore, we provide measurements on these features as default
measurements. The measurement methods are typically classified into three approaches:
tape-measurement, contour-measurement, and linear-measurement. Tape-measurement is 13
done traditionally by stretching a leather tape across the concave portion of the body to
obtain body circumferences. The length from contour-measurement is the distance between
two landmarks along the body surface. The method is similar to tape-measurement except
it follows the surface of the body without spanning across the concave parts. The height or
breadth of two landmarks is measured by the linear distance between two body landmarks.
In this research, the convex-hull method is applied to mimic manual
tape-measurement. Both lengths obtained by tape-measurement and
contour-measurement share the same integration operation of piecewise deviation. In
frontal and lateral perpendicular projections of the meshing surface, we are able to
obtain linear measurements by a simple calculation in vertical or horizontal distance.
Here, we give a detailed description of the tape-measurement method.
In practice when taking tape-measurement, the tape only contacts the convex part of
the body to form a convex hull around the body. For a co-planar feature, e.g. the bust
girth shown in Figure 5, the convex hull may be visualized as an imaginary elastic
band stretched open to enclose the body; when released the band forms the shape of
convex hull around the body.
Tape-measurement is classified into two types: closed circumference and open-end
curve. First, determine the measurement position with respect to the landmarks on the
structure mesh. For closed circumference tape-measurement, a cutting plane CP is
defined by three landmarks on the mesh surface. A curve C is obtained by finding the
intersection between structure mesh and the cutting plane CP. For open-end
tape-measurement, for example, the blue solid line depicts the corresponding feature
line of outside leg line, shown in the red curve in Figure 5. And the red curve shown in
Figure 6 is the outside leg line drawn on the structure mesh. In order to span the
concave portion, a virtual point V0 is added inside the body to connect two ends of

Figure 5.
Convex hull for bust girth
tape-measurement
IJCST
25,1

14

Tape-measurement

V0 Outside leg line

Figure 6.
Convex hull of
outside leg line

the feature line to form a closed polygon as curve C. Therefore, the length of the
open-end tape-measurement is simply approximated by subtracting the lengths of two
dash lines. The convex hull Hconvex(P) of curve C is given by:

(  )
X
k  Xk

H convex ðPÞ ¼ ai pi  pi [ C; ai [ R; ai $ 0; ak ¼ 1; k ¼ 1; 2. . . ð6Þ
i¼1
 i¼1

A rough estimation of the measurement can be given by summing the inter-vertex

distance of Hconvex(P):

X
n21
Length ¼ kpi 2 piþ1 k; pi [ H convex ðPÞ ð7Þ
i¼0
However, precision of the linear distance is poor unless the resolution of the mesh surface A feature-based
is enhanced. But, how dense it should be to achieve a satisfied precision? In order to
obtain a good estimation of the measurement with a sparse mesh, we employed a
anthropometry
B-spline approximation technique (Li, 2004) to regenerate the surface curve. For
reasonable curve smoothing, we select the frequently used order k ¼ 4 for B-spline C(u):
X
n
CðuÞ ¼ N i;k ðuÞP i ; u [ ½tk21 ; tnþ1 Þ ð8Þ 15
i¼1

where Pi is the control points, and Ni,k(u) is the kth-order B-spline basis functions. Thus,
the minimized square distance approximation is given by:
X
m
2
E¼ kd i 2 Cðui Þk ð9Þ
i¼1

Therefore, the control points of the approximated B-spline is obtained by:

½P
¼ ½M
21 ½Q
ð10Þ

 Pm  Pm 
where M ¼ i¼1 N i;k ðui ÞN j;k ðui Þ and Q ¼ i¼1 d i N i;k ðui Þ .
By solving equation (10), the approximation curve is obtained. The length of the
curve (i.e. the measurement of a feature) is given by numerical integration:
Z u¼1
Length ¼ c0 ðuÞdu ð11Þ
u¼0

4. Evaluation
Here, we present two approaches of software accuracy evaluation by comparing
measurement deviations on different mathematical methods and the effect on accuracy
by noise. By using B-spline and polygon to approximate the body circumference for
measurements, we found that B-spline approximation is fairly successful for body
measurements. Meanwhile, the method we employed for digitalanthropometry
calculation is relatively insensitive to sample frequency and noise.
Shapes of human body circumferences are circle-like closed curves. Therefore, in the
first part of the test, we evaluate the circumference of a circle in different sampling by two
mathematical methods. The first one is B-spline approximation, and the second is the
summation of all piecewise linear segments. With increasing number of sampling points,
we can anticipate that the result will converge toward the circle circumference. Since we
know the value of the circle circumference, we are able to assess the errors generated by
different calculation methods. The convergent curves are drawn in Figure 7(a). The
horizontal axis denotes the number of sampling point, and the vertical axis denotes the
accuracy of the circumference. The accuracy of the B-spline approximation is noticeably
better than piecewise linear summation. Figure 7(b) shows a similar test on a digital
human mesh model. By investigating the waist girths in different mesh resolutions, it was
discovered that the convergence speed by B-spine approximation is very good while a
large amount of mesh points are needed to evaluate the circumferences of the subject.
The second test is the evaluation of the degree of noise sensitivity of these two
methods. Noises are added to the sample circle data to mimic noises that are usually
IJCST
25,1

16

(a)

Figure 7.
(a) Sample numbers
needed to obtain an
acceptable accuracy,
(b) the waist circumference
calculated by different
mesh resolution and
(c) the effect of noise
applied on the accuracy of
circumference
(b) (c)

generated by the scanner. The level of perturbation ranges from 1 to 5 per cent of the
circle radius. The circumference measurement is drawn in Figure 7(c). The horizontal
axis denotes the percentage of noise added into the sample data, and the vertical axis
denotes the percentage of error with respect to the circle circumference. We found that
errors produced from the B-spline approximation method is generally less than
50 per cent of the level of noise added to the sampling points. In regard to the polygonal
method, the summation of linear segments is very sensitive to noise; the error of the
circumference exceeds the level of perturbation when the noise increases. Based on the
outcomes of the tests, we conclude that B-spline curve reconstruction for body
measurement is relatively insensitive to sample frequency and noise.

5. Results
These mesh modeling and measurement operations are developed using the
C þ þ programming language and OpenGL graphics library. All results are tested on
a Pentium IV 3.0GHz personal computer. It takes less than one minute to complete a
structure modeling and 57 standard measurements starting from the scanning cloud
points. Figure 8(a) shows the integrated measurement and display environment. The right
half of the window shows the body mesh and the overlaying feature lines. The user can
examine the level of detail of the reconstructed mesh. Users can also toggle individual
feature options to show in the right window. Figure 8(b) shows the enlarged view of the
upper left window; every default feature can be displayed one by one as representative
 A feature-based
anthropometry

17

(a) (b)

Figure 8.
(a) The user interface of
the measurement system,
(b) enlarged window of
cross section view, (c) list
of torso measurements
and (d) list of extremities
measurements
(c) (d)

interior nodes and its bounded measurement curve. In addition, it allows us to show the
other cross-sections of the body by setting the height a mobile horizontal plane. The
detailed information including the circumference and height is listed in it. The lengths of
the most commonly used measurements are listed on the lower left window. The features
are classified into two pages based on their location of the body, shown in Figure 8(c) and
(d). Each feature is listed by name, followed by the type of measurement and the digital
measurement data. In the cases of a feature that is present on both sides of the body, the
right side body measurement is enumerated on the fourth column.
Several Asian female subjects of different ages and distinct figures were scanned and
tested in the developed software. Each subject was scanned three successive times;
between each scan, the subject was asked to reposition herself in the scanner to evaluate
the effect of positioning on measurement consistency. Table I lists summary
measurements of the subjects. Each measurement is followed by a superscript letter
IJCST 001 002 003 004 005 Mean SD
25,1
Age (years) 47 45 72 31 20 43 19.6
Weight (kg) 49 57 52 57 44 52 5.5
Stature (cm) 158 158 153 161 150 156 4.4
Mid-neck girthT 201 237 207 197 186 206 19.4
Neck base girthT 230 235 234 216 193 222 17.8
18 Upperchest girthT 502 549 538 489 440 503 43.3
Chest girthT 494 596 588 480 541 540 52.6
Underbust girthT 401 420 465 404 331 404 48.4
Waist girthT 375 406 423 395 298 379 49.0
Hip-hip girthT 444 456 511 493 349 451 63.0
Hip girthT 451 482 518 491 378 464 53.8
Shoulder height (left)L 1,315 1,301 1,297 1,338 1,208 1,292 49.6
Shoulder height (right)L 1,314 1,295 1,296 1,336 1,196 1,287 53.8
Crotch heightL 718 666 699 746 660 698 35.9
Chest depthL 220 241 262 221 201 229 23.3
Chest breadthL 270 296 295 302 245 282 23.7
Hip breadthL 330 361 347 374 307 344 26.1
Across shoulderT 370 347 374 392 388 374 17.8
Armhole (left)T 202 265 324 212 219 244 50.5
Armhole (right)T 247 273 334 270 188 262 52.7
True riseL 292 312 271 277 275 285 16.7
DropT 119 190 165 85 243 161 61.5
Front high-hipC 303 301 329 350 263 309 32.9
Scye depthL 164 200 225 178 187 191 23.3
Shoulder girthT 221 256 278 239 195 238 31.6
Shoulder length (left)C 132 138 139 158 141 141 9.8
Shoulder length (right) C 140 139 137 151 146 143 5.8
Back widthC 304 285 297 356 302 309 27.2
Total crotch lengthC 606 660 577 587 567 599 36.8
Thorax depth at nippleL 220 241 261 220 201 229 23.1
Center back waist lengthC 193 231 174 202 209 202 21.0
Center front waist lengthC 186 223 170 192 205 195 19.8
Cervicale to bust pointC 311 323 346 317 324 324 13.3
Cross chest widthC 314 324 326 334 282 316 20.1
Total vertical trunkT 1,234 1,315 1,260 1,246 1,141 1,239 63.0
Thigh girth (left)T 346 359 320 410 319 351 37.0
Thigh girth (right)T 358 358 338 368 295 343 29.2
Calf girth (left)T 210 238 202 225 186 212 20.0
Calf girth (right)T 203 229 195 202 181 202 17.5
Ankle girth (left)T 110 117 128 120 109 117 7.8
Ankle girth (right)T 114 119 120 119 121 119 2.9
Ankle height (left)L 109 100 122 115 85 106 14.2
Ankle height (right)L 113 104 124 117 87 109 14.5
Knee girth (left)T 221 244 232 225 189 222 20.4
Knee girth (right)T 216 231 221 210 179 212 19.5
Outside leg length (left)T 979 973 950 1,012 943 971 27.4
Outside leg length (right)T 982 978 958 1,030 941 978 33.7
Table I. Mid-thigh girth (left)T 278 283 258 309 211 268 36.4
The default Mid-thigh girth (right)T 280 270 252 302 210 263 34.6
measurements for Wrist girth (left)T 73 90 88 76 82 82 7.7
women subjects (continued)
 A feature-based
001 002 003 004 005 Mean SD
anthropometry
Wrist girth (right)T 77 92 97 77 82 85 9.3
Arm length (left)C 505 512 507 524 508 511 7.6
Arm length (right)C 525 505 535 528 491 517 18.3
Elbow girth (left)T 137 154 183 169 121 153 24.5
Elbow girth (right)T 144 166 184 158 124 155 22.6 19
Underarm length (left)C 245 243 225 249 243 241 9.1
Underarm length (right)C 260 236 246 257 256 251 9.9
Upperarm girth (left)T 163 177 161 185 141 165 17.0
Upperarm girth (right)T 157 161 152 158 133 152 11.3
Upperarm length (left)C 263 276 297 280 262 276 14.1
Upperarm length (right)C 267 272 300 274 236 270 22.9
Note: Unit: mm Table I.

denoting the type of the measurement. We use T for tape-measurement, C for

contour-measurement, and L for linear-measurement. The data listed in the table is the
result of three scans; the variations between the three scans are listed in Table II. The last
two columns demonstrate the mean and standard deviation (SD) among these five
subjects.
In this study, we used the mean absolute difference (MAD) and the coefficients of
variation (CV) to quantify the consistency between scans. Table II lists the variations
of each individual subject. The MAD is the average absolute deviation from the mean.
We use the allowable error proposed by ANSUR (Gordon et al., 1989) as a standard to
evaluate the consistency of the anthropometric system. For the measurements that are
not list the in ANSUR, we use the strict accuracy of ^ 5 mm defined in ISO-8559 for
consistency evaluation in our method. We found that the MAD values of all but one
measurement are less than the error allowance. One of the worst cases, a drop, which
defines the difference between the chest and waist, is within the boundaries of
acceptable allowance.
The CV is another common measurement of consistency, which is defined as the SD
divided by the mean. In general, a CV below 5 per cent (Robinette and Daanen, 2006)
represents a consistent system. Of 57 measurements in this study, 44 CVs are less than
2 per cent while four are less than 3 per cent.

6. Conclusion and discussion

In this paper, we propose an automatic anthropometry method by transforming body
scan data into a meaningful digital mannequin with embedded feature-based structure.
The size of the scanned data can be reduced to hundredths of its original size as long as
the required precision is preserved. The data structure of the digital mannequin is
respectably compact and significant such that it includes all body features. Every
geodetic coordinate carries its own geometrical meaning that represents a specific
feature location. By using the procedure of mesh development, one can easily build a
human-shaped mannequin with different levels of detail while still retaining accurate
anthropometric features.
 20
25,1
IJCST

Table II.
Variations of
measurements
CV(%) MAD(mm) ANSUR
Measurements 1 2 3 4 5 Mean 1 2 3 4 5 Mean Error allowance (mm)

Mid-neck girthT 2.2 1.8 0.3 0.3 2.1 1.3 3.3 3.1 0.4 0.5 2.8 2.0 6
Neck base girthT 1.4 1.7 1.5 1.4 1.8 1.6 2.4 3.1 2.7 2.2 2.4 2.6 11
Upperchest girthT 1.8 0.3 0.6 1.2 0.8 0.9 6.9 1.1 2.1 4.3 2.6 3.4 15
Chest girthT 0.6 0.4 0.8 1.2 1.0 0.8 2.0 1.9 3.4 4.6 3.9 3.1 15
Underbust girthT 1.0 1.2 0.8 0.7 1.1 1.0 2.9 3.8 2.7 2.0 2.9 2.8 16
Waist girthT 0.5 0.7 2.0 1.3 1.4 1.2 1.3 1.9 5.9 3.9 2.8 3.2 11
Hip-hip girthT 0.3 0.5 1.2 1.1 0.6 0.7 0.9 1.4 4.4 3.9 1.4 2.4 12
Hip girthT 0.6 1.1 0.1 0.6 0.2 0.5 1.8 4.1 0.2 2.1 0.6 1.8 12
Shoulder height (left)L 0.2 0.1 0.2 0.1 0.1 0.1 1.8 0.8 2.1 1.1 0.5 1.3 7
Shoulder height (right) L 0.1 0.3 0.4 0.2 0.6 0.3 0.7 2.5 3.8 2.2 4.8 2.8 7
Crotch heightL 0.8 0.8 1.0 1.0 0.7 0.9 4.2 3.9 4.8 5.0 3.2 4.2 6
Chest depthL 1.5 1.3 0.4 0.7 1.3 1.0 2.4 2.4 0.8 1.2 1.9 1.7 4
Chest breadthL 3.2 2.4 1.3 1.6 0.7 1.8 6.0 5.5 2.8 3.6 1.2 3.8 10
Hip breadthL 1.2 1.7 0.4 0.5 1.1 1.0 2.9 4.1 1.1 1.5 2.6 2.4 7
Across shoulderT 0.3 0.5 1.5 1.3 0.4 0.8 0.7 1.3 4.3 4.0 1.3 2.3 N/A
Armhole (left)T 3.5 5.1 2.3 2.1 3.9 3.4 5.3 9.1 5.3 3.1 6.4 5.9 13
Armhole (right)T 3.4 4.0 4.2 3.1 4.7 3.9 5.8 7.7 9.9 6.1 5.9 7.1 13
True riseL 1.0 3.4 2.0 2.9 1.0 2.1 2.0 8.2 3.9 5.6 2.1 4.4 N/A
DropT 3.9 5.4 5.0 2.7 3.4 4.1 3.3 7.5 6.3 1.6 6.4 5.0 N/A
Front high-hipC 1.1 1.2 1.0 3.2 1.3 1.6 2.4 2.7 2.1 8.4 2.6 3.7 6
Scye depthL 3.9 2.0 1.1 1.4 3.2 2.3 4.9 3.1 1.8 1.8 4.2 3.2 4
Shoulder girthT 1.2 0.6 0.6 1.8 1.5 1.1 2.0 1.2 1.2 3.3 2.1 1.9 22
Shoulder length (left)C 2.0 1.8 0.8 0.6 3.6 1.8 2.0 1.8 0.9 0.7 3.8 1.8 3
Shoulder length (right)C 1.1 3.0 1.2 1.2 1.7 1.6 1.1 3.1 1.1 1.4 1.8 1.7 3
Back widthC 2.0 1.3 1.4 0.7 2.0 1.5 4.7 2.9 3.1 1.8 4.3 3.4 10
Total crotch lengthC 0.6 3.3 1.8 2.8 1.1 1.9 2.7 16.7 7.4 11.7 4.6 8.6 16
Thorax depth at nippleL 1.5 1.1 0.7 1.2 1.1 1.1 2.4 2.1 1.2 1.9 1.7 1.9 4
Center back waist lengthC 4.4 2.7 2.1 1.5 1.2 2.4 6.4 4.5 2.8 2.4 1.9 3.6 N/A
(continued)
 CV(%) MAD(mm) ANSUR
Measurements 1 2 3 4 5 Mean 1 2 3 4 5 Mean Error allowance (mm)

Center front waist lengthC 5.0 3.0 2.4 1.4 1.3 2.6 7.1 4.8 2.9 1.8 2.1 3.8 N/A
Cervicale to bust pointC 1.5 0.4 1.7 0.4 1.1 1.0 3.3 1.0 4.4 0.8 2.5 2.4 8
Cross chest widthC 1.5 0.5 0.8 0.2 1.4 0.9 3.6 1.3 2.0 0.4 2.9 2.0 N/A
Total vertical trunkT 0.3 0.7 0.7 1.0 0.4 0.6 2.7 6.5 7.2 8.9 3.0 5.7 22
Thigh girth (left)T 0.2 0.7 1.8 1.4 0.5 0.9 0.6 1.8 4.1 4.1 1.1 2.3 6
Thigh girth (right)T 0.6 1.1 1.1 0.7 1.3 1.0 1.4 3.1 2.7 1.8 2.8 2.3 6
Calf girth (left)T 1.5 2.0 2.2 2.0 2.5 2.0 2.4 3.6 3.3 3.3 3.5 3.2 5.0
Calf girth (right)T 1.7 2.7 0.9 1.0 1.6 1.6 2.5 4.4 1.2 1.4 2.2 2.3 5.0
Ankle girth (left)T 1.6 2.0 2.0 2.1 2.6 2.0 1.3 1.6 2.0 1.8 2.0 1.7 4
Ankle girth (right)T 1.3 3.3 2.7 2.1 1.1 2.1 1.1 3.0 2.5 1.8 1.0 1.9 4
Ankle height (left)L 3.3 1.4 4.5 0.5 2.3 2.4 2.7 1.1 3.6 0.5 1.5 1.9 3
Ankle height (right)L 1.3 3.3 1.6 3.3 0.6 2.0 1.0 2.7 1.5 2.6 0.3 1.6 3
Knee girth (left)T 1.8 1.1 1.8 0.1 1.6 1.3 3.0 1.8 3.3 0.1 2.0 2.1 4
Knee girth (right)T 1.5 2.1 2.7 1.2 1.1 1.7 2.4 3.2 4.3 1.7 1.5 2.6 4
Outside leg length (left)T 0.2 0.6 0.6 0.3 0.5 0.4 1.1 4.0 3.8 2.6 3.3 2.9 N/A
Outside leg length (right)T 0.3 0.2 0.2 0.2 0.3 0.2 2.1 1.4 1.5 1.3 2.5 1.8 N/A
Mid-thigh girth (left)T 0.3 1.4 0.7 1.1 1.0 0.9 0.6 3.0 1.4 2.5 1.5 1.8 4
Mid-thigh girth (right)T 1.2 0.4 0.9 0.9 1.9 1.1 2.5 0.8 1.7 1.8 2.8 1.9 4
Wrist girth (left)T 3.7 1.9 4.4 3.0 6.0 3.8 1.9 1.3 3.0 1.7 3.5 2.3 11
Wrist girth (right)T 3.3 1.2 1.6 2.6 1.6 2.0 1.9 0.8 1.2 1.5 1.0 1.3 11
Arm length (left)C 1.2 0.6 0.7 1.2 1.0 0.9 4.2 2.2 2.7 4.3 3.7 3.4 6
Arm length (right)C 0.7 0.4 0.6 0.2 0.2 0.4 2.7 1.5 2.5 0.7 0.7 1.6 6
Elbow girth (left)T 1.9 3.7 2.4 1.0 4.9 2.8 1.8 4.0 3.2 1.3 4.0 2.9 4
Elbow girth (right)T 2.0 2.4 1.5 2.6 0.8 1.9 2.2 3.1 2.1 3.1 0.7 2.2 4
Underarm length (left)C 1.3 1.2 2.7 1.1 1.4 1.6 2.3 2.0 4.1 2.1 2.7 2.6 6
Underarm length (right)C 2.0 2.5 1.2 0.7 1.5 1.6 3.7 3.9 2.3 1.2 2.6 2.7 6
Upperarm girth (left)T 1.5 1.8 1.3 0.6 1.8 1.4 1.9 2.5 1.6 0.8 1.8 1.7 6
Upperarm girth (right)T 2.3 1.3 1.0 2.8 1.7 1.8 2.5 1.5 1.0 3.4 1.8 2.0 6
Upperarm length (left)C 1.0 1.3 1.3 0.2 2.0 1.2 2.0 2.5 2.9 0.3 3.8 2.3 6
Upperarm length (right)C 0.9 1.7 1.9 1.9 2.2 1.7 1.7 3.4 4.0 3.7 3.8 3.3 6
anthropometry
A feature-based

21

Table II.
IJCST The benefit of the embedded information in the structure mesh, detail
25,1 and precise measurements is that the data can be taken in a very short period of time.
The most frequently used measurements in apparel industries are demonstrated in this
study. In addition, additional feature curves that pass through several feature points
defined on the mesh structure can be defined by users for their individual needs.
In the research, we were able to measure all features in a standing pose within the
22 acceptable allowance proposed by ANSUR. When comparing the reliability of the three
types of measurements mentioned previously, the MADs of linear measurements are
generally smaller since they are not obstructed during the scan and are primarily
derived by the position of landmarks. Even so, the MADs of the tape- and
contour-measurements are less than standard allowance. The worst MAD values
obtained in this research are the circumferences of the armholes, total crotch length, and
drop. The MAD values of these measurements range from 5.0 to 8.6 mm and are greater
than the 5 mm threshold proposed by ISO 8559. The variations are still below the
ANSUR allowable error, though. The cause of the greater errors of the former three
measurements is due to a smaller amount of data collected from scanner; thus recreating
meshes of armholes and crotch regions in perfect geometry is almost impossible. Local
additional scans for the obstructed regions can improve the outcomes of the evaluation.
The reason that the variation of the drop is greater than that of other measurements is
the fact that it is derived from two independent features. Thus, its error is magnified by
the variations of its associated underlying components.
This research provided an important infrastructure for subsequent development of
automatic anthropometry system incorporated with ordinary body scanners. Based on
the successful outcomes of body feature identification, we can predict that tedious and
labor-intensive national sizing data will be much easier to collect.

References
Ashdown, S. and Loker, S. (2001), “Use of body scan data to design sizing systems based on
target markets”, America National Textile Center Annual Report, National Textile Center,
Blue Bell, PA.
ASTM (1999), Standard Terminology Relating to Body Dimensions for Apparel Sizing, ASTM
Designation: D 5219-99, ASTM, West Conshohocken, PA.
Carrere, C., Istool, C., Little, T., Hong, H. and Plumlee, T. (2000), “Automated garment
development from body scan data”, National Textile Center Annual Report: I00-S15,
November 2000, 2001, National Textile Center, Blue Bell, PA.
Cyberware (2005), available at: www.cyberware.com
Dekker, L., Douros, I., Buston, B.F. and Treleaven, P. (1999), “Building symbolic information for
3D human body modeling from range data”, Proc. of Second International Conference on
3D Digital Imaging and Modeling, pp. 389-97.
Gordon, C.C., Bradtmiller, B., Clausen, C.E., Churchill, T., McConville, J.T., Tebbetts, I. and
Walker, R.A. (1989), 1987-1988 Anthropometric Survey of US Army Personnel: Methods
& Summary Statistics, Natick/TR-89-044, US Army Natick Research Development and
Engineering Center, Natick, MA.
Inspeck (2005), available at: www.inspeck.com/
ISO (1989), Garment Construction and Anthropometric Surveys – Body Dimensions, Reference
No. 8559-1989, ISO, Geneva.
Jones, R.M., Li, P., Brooke-Wavell, K. and West, G.M. (1995), “Format for human body modeling A feature-based
from 3D body scanning”, International Journal of Clothing Science & Technology, Vol. 7
No. 1, pp. 7-16. anthropometry
Ju, X., Werghi, N. and Siebert, J.P. (2000), “Automatic segmentation of 3D human body scans”,
Proc. of IASTED International Conference on Computer Graphics and Imaging 2000
(CGIM 2000), Las Vegas, NV.
Leong, I.F., Fang, J.J. and Tasi, M.J. (2007), “Automatic body feature extraction from a 23
marker-less scanned human body”, Computer-Aided Design, Vol. 39 No. 7, pp. 568-82.
Lin, C.C. (2003), “A study on the development of a body scanner and processing of the scanned
data”, Master dissertation, National Cheng Kung University, Tainan.
Nurre, J.H. (1997), “Locating landmarks on human body scan data”, Proc. of International
Conference on Recent Advances in 3-D Digital Imaging and Modeling, pp. 289-95.
Pargas, R.P., Staples, N.J. and Davis, J.S. (1997), “Automatic measurement extraction for apparel
from a three-dimensional body scan”, Optics and Lasers in Engineering, Vol. 28, pp. 157-72.
Robinette, K.M. and Daanen, H.A.M. (2006), “Precision of the CAESAR scan-extracted
measurements”, Applied Ergonomics, Vol. 37 No. 3, pp. 259-65.
Robinette, K., Boehmer, M. and Burnsides, D. (2001), “3-D landmark detection and identification
in the CAESAR project”, Proc. 3rd International Conference on 3-D Digital Imaging and
Modeling, pp. 393-8.
Robinette, K., Daanen, H. and Paquet, E. (1999), “The CAESAR project: a 3-D surface
anthropometry survey”, Procs. of IEEE Second International Conference on 3-D Digital
Imaging and Modeling, pp. 380-6.
TC2 (2005), available at: www.tc2.com/
TriForm (2005), available at: www.wwl.co.uk/
Tsai, M.J., Hwang, H.R., Lin, C.C. and Chen, A.P. (2003), “A 3D body scanner”, Taiwan Patent
No. 203020.
Vitus (2005), available at: www.vitronic.com/
Wang, C.L., Chang, K.K. and Yuen, M.F. (2003), “From laser-scanner data to feature human
model: a system based on fuzzy logic concept”, Computer-Aided Design, Vol. 25 No. 3,
pp. 241-53.
Zhong, Y. and Xu, B. (2006), “Automatic segmenting and measurement on scanned human
body”, International Journal of Clothing Science & Technology, Vol. 18 No. 1, pp. 19-30.

Corresponding author
Jing-Jing Fang can be contacted at: [email protected]

To purchase reprints of this article please e-mail: [email protected]

Or visit our web site for further details: www.emeraldinsight.com/reprints

View publication stats