2nd IEEE International Conference on Computational Systems and Information Technology for Sustainable Solutions 2017

IoT enabled Video Surveillance System using

Raspberry Pi
Rahul Muppanagouda Patil1, Ram Srinivas2, Rohith Y3, N R Vinay4 and Pratiba D5
Department of Computer Science and Engineering
R.V. College of Engineering, Bangalore
Karnataka, India
Email: {1rahulpatil61197, 2ramsrinivas57, 3rohithgoud10, 4nrvinay08}@gmail.com, [email protected]

Abstract—The Internet of Things (IoT) is the there is a constant need by users to monitor people and entities
communication of various devices or “things” with each that are dear to them and not everybody is capable of meeting
other by the sharing of usable and relevant data. In the high prices and costs of these systems. Therefore, with the
today’s age of digitization and “smart” devices, one of the introduction of small and smart standalone computers such as
foremost requirements is for new, innovative technology the Raspberry Pi, any user can develop their own surveillance
integral to the concepts of Internet of Things in everyday systems that are not just isolated systems, but ones that also
life. The need for surveillance and monitoring has interact with various other devices and infrastructures.
drastically increased over the course of the past few Businesses can also provide such products at much lower
decades. Both private areas and public areas alike require costs now, very much in contrast to the systems provided in
these types of surveillance systems to ensure security, and the past which were both costly and also required a lot of
are generally very costly. In this paper, is proposed, an manual monitoring.
innovative surveillance system which is powered by the
Raspberry Pi, Amazon Web Services and Google Drive
that provided results with minimal latencies.

Keywords— Motion Detection, OpenCV, AWS, Raspberry Pi,

Video Surveillance, Internet of Things, Google Drive

I. INTRODUCTION

The Internet of things (IoT) is the network of physical

devices such as automobiles, home appliances, and other
items embedded with electronics, software, sensors, actuators,
and network connectivity which enable these objects to
connect and exchange data. Each thing is uniquely identifiable
through its embedded computing system but is able to inter-
operate within the existing Internet infrastructure. The concept
of IoT is currently being used in several fields such as
agricultural industry, automotive industry, medical industry
and of course, in security and surveillance. IoT expects use of
low cost consuming and low power consuming devices with
controlled adverse effects on the environment.
Fig. 1. Surveillance System Architecture
Video Surveillance systems have been an extremely
essential part of security systems installed in all types of Applications such as Video Surveillance have become
offices, schools, hotels, businesses and even people‟s homes. palpably better by their incorporation in the Internet of Things
The traditional methods of having around-the-clock system, which is merely just connecting several devices and
surveillance using CCTV systems bore great burdens in the transferring relevant data from one to another, preferably in
form of costs and maintenance to the consumer. They also real-time.
lacked communication with other devices and were restricted
to the sphere in which the system existed. The need of a In this paper, is further elaborated, a cost-effective
connected surveillance system is higher than ever today, as surveillance system that has been developed using the popular

978-1-5386-2044-1/17/$31.00 ©2017 IEEE 308

 2nd IEEE International Conference on Computational Systems and Information Technology for Sustainable Solutions 2017

background subtraction method to detect motion. The system time. However, existing systems are not always area or room-
also incorporates various other features using services specific, and do not alert the user about an infiltration,
provided by Amazon Web Services, such as SES for notifying irrespective of the user‟s location. In addition, none are
the user via e-mail and AWS RDS for storing relevant data. implemented using systems akin to the Raspberry Pi.
Google drive has been used to store images on cloud to make
it easily accessible to the user. A client-side GUI for users has
IV. IMPLEMENTATION
also been developed, to view data in a user-friendly,
interactive format and also to gain relevant insights from
them. Hence, an IoT based system has been successfully The system uses three-tier architecture, where the first tier
implemented using the Raspberry Pi, AWS as well as Google is the background subtraction algorithm running to detect
Drive. motion that keeps constantly running in a loop triggering
certain functions when deemed necessary. The second tier is
the database and cloud storage which stores important data in
II. RELATED WORK cloud servers. It also includes the e-mail servers that aid in
notifying the user in real-time. The third and final tier is the
part of the system that experiences user interaction. This
Extensive research in the field of computer vision has includes a web application that has been developed so that the
paved way for many successful systems that utilize image user can observe information related to their system.
processing techniques. The most common techniques include
background subtraction and differential frame analysis. [1] The system is essentially divided into various steps or
proposes a three layer architecture. The first layer is the processes as listed below:
motion detection layer, which is the outcome of a Python
script and the second layer triggers the actions based on a 1. Video feed is continuously transferred from USB
configuration file. The third layer triggers actions like sending Camera to Raspberry Pi.
an email, storing the images/videos to an FTP server. Live 2. Motion detection script constantly running the
video streaming system based on Raspberry Pi connected to a background uses these input frames to make a
5MP Camera Module. The sensitivity of their system [2] decision based on certain parameters.
revolves around the chosen value for a threshold which is a
measure of the rate of change of a pixel location. [3] delivers a 3. If motion is detected, and is long enough to be
system that compares two successive images and determines regarded as a significant motion, the frame in which
the motion and takes the necessary steps following the motion is detected is immediately uploaded to the
detection. The method proposed in [4] uses a combination of drive in cloud.
temporal frame differences and background subtraction which 4. User is notified with relevant information such as
can successfully detect slow moving targets as well. To motion start time along with an image URL that links
improve the accuracy, a cross-correlation method is adopted to the image uploaded to the drive in previous step.
for better results regarding slow and stationary objects and
background updating is employed. A three-tier based approach 5. Motion time is uploaded into the RDS in cloud for
was proposed in [5], one for efficiently determining the future analysis and visualization.
background (BM Module) using rapid matching followed by 6. User can log into a web application using given
accurate matching, second being a novel block based entropy credentials to view important information regarding
evaluation method for identification of block candidates of their systems and also visualize the data uploaded to
moving targets, and third being the object extraction (OE database.
Module) which examined every block to identify a moving
object.
The system proposed in this paper focus primarily on the
intercommunication of devices and services to build a
successful working surveillance system. It is highly capable
and will work with any of the latest motion detection
algorithms that have been proposed up till today.

III. MOTIVATION

Existing systems that are commonly in use monitor areas,

leaving their footage to be reviewed at a later stage.
Alternatively, there exist many common security systems that
serve as alert systems upon the intrusion of an infiltrator via
one or many entry points, mostly to serve as a security system
for residents residing within an apartment or building at the Fig. 2. Three tiers of the System

309 978-1-5386-2044-1/17/$31.00 ©2017 IEEE

 2nd IEEE International Conference on Computational Systems and Information Technology for Sustainable Solutions 2017

Fig. 4. Background Subtraction

The algorithm has been implemented in Python using the

OpenCV module. The methods used on successive frames to
detect motion are as described below:

1) First each frame is captured by the camera and

transferred to the Raspberry Pi. If the frame is valid,
then it is used for upcoming operations.
2) The frame is resized to half the size to reduce
computations and is then converted to a single channel
image. This makes finding the absolute difference
Fig. 3. System Flowchart between frames easy and less expensive.
3) The image is blurred using Gaussian Blur to de-noise
the image and is then fed to the accumulateWeighted
A. Motion Detection using Background Subtraction function to model the background dynamically.
The primary function of the system is to detect motion 4) The absolute difference is calculated between the
that occurs in a video stream,which is being captured and fed current frame and the dynamic background to obtain
as input continuously. The method used must be able to make Fig. 4.1 in which motion can be observed.
the decision in real-time and must therefore be quick and 5) This image is then applied to a threshold so that only
accurate. Motion detection using background subtraction is significant changes are considered, and is dilated to
essentially performed by just comparing successive frames, form proper solid entities that depict entity that caused
and if significant change is noticed then motion is said to be the motion.
occurred. 6) Finally contours are found in this dilated image, and
Also, the duration of the detected motions is considered only if contours are larger than a certain pre-defined
by counting the number of frames in which motion is detected. threshold then a motion is said to be detected.
Only if the number of these frames is greater than a minimum 7) Finally a bounding rectangle is drawn over the entity
threshold, a response will be triggered. that caused the motion by obtaining the coordinates of
The system also provides a provision where in motions the contour and passing them as parameters to draw the
that occur with minimal time gap be considered as one single rectangle.
motion. The time elapsed between successive motions is
calculated for every motion detected and is considered This algorithm will tend to give inaccurate results in low
different from the preceding detected motion only if the actual light conditions. It detects changes in lighting as motion and
time elapsed is greater than or equal to some minimum time does not perform to its optimum when there are shadows in
gap as specified by either the user or developer. This time gap the video. It also detects non-threatening motion such as
can vary according to what type of environment the displacement of household objects, which are nothing but
surveillance system is used for. false results.

978-1-5386-2044-1/17/$31.00 ©2017 IEEE 310

 2nd IEEE International Conference on Computational Systems and Information Technology for Sustainable Solutions 2017

B. Storing image and relevant data on Cloud server detected. It also contains a link to the image uploaded into the
In the Internet of Things era there is a need for this type Google drive for that corresponding motion.
of surveillance system to have the obtained data to be
accessible and available to other authorized systems and D. Data visualization and User interaction
applications, and not just be stored locally in the Raspberry Pi.
For the application to be wholesome, it is imperative that
The system satisfies this need by using a popular AWS service
the user functionality extend beyond merely that of receiving
called RDS (Relational Database System) which is hosted on a
e-mail alerts and viewing pictures. There is a need for the user
virtual instance in the cloud. This remote storage makes it
to be able to view all the data over a certain time-frame in
readily accessible to other applications that can give really
some format, from a remote location at his/her leisure. Hence,
good deductions from the data. It also helps the user to view
a web portal was developed for the same. The user is required
information obtained from their systems in real-time from any
to merely login to the portal using his/her assigned credentials,
authorized device that has access to the network.
and can subsequently view all the stored data regarding
The system uses the PyMySql library to establish foreign movements detected in the form of various interactive
connection to the MySQL server in the cloud and enables graphs. The user could even choose to view the stored data in
execution of SQL queries to insert or update data in the table. raw text form.

To store images of the instances when motions are Python‟s Flask module was used to develop the web
detected, popular cloud storage service „Google Drive‟ is portal, along with HTML and CSS for the UI, while another
used. This is done using the PyDrive API provided by Google couple of Python libraries, Bokeh and Pandas are used to plot
to authenticate the user, upload image to drive and to extract a and render the interactive graphs from which the user can infer
shareable link to the image uploaded directly from our main a lot of information such as rate of occurrences of motion,
program. average duration of motions, etc.

V. E XPERIMENT RESULTS AND A N ALY SIS

The system is developed to detect motion in real-time and

respond to significant motions appropriately by notifying the
user immediately about this anomaly and logging important
information such as start and end of detected motion times. It
also uploads and stores the image frame in which motion was
detected.

Fig. 5. Motion time and image URL entries in AWS RDS

C. Notifying the User

Simple Email Service is a free service provided by
Amazon Web Service. This is part of our second tier which
alerts the user if any motion were to occur. AWS has provided
a Python module called boto3 which enables any AWS
account holder to access all the services directly from their
Python scripts.

On a Linux System, boto3 can be installed by using the

pip package in Python: sudo pip install boto3 and the
credentials can be setup by specifying certain parameters such
as the access key id, secret access key and the region in which
the instance is hosted. Fig 6. Motion with single entity

This feature is used to access the cloud services such as Results of earlier experiments with movement caused by
Amazon SES that notifies the user via e-mail, where the email a single entity are as shown in the collection of figures in Fig.
contains important information such as the time motion was 6 which depicts the entry of an entity from the bottom of the

311 978-1-5386-2044-1/17/$31.00 ©2017 IEEE

 2nd IEEE International Conference on Computational Systems and Information Technology for Sustainable Solutions 2017

frame and gradually moving upwards. The region of detection At the end of the motion, the end time of the motion is
of the motion has been highlighted by a bounded green also logged into the RDS as it helps in identifying various
rectangle. important factors in the future.
On detection of the motion, the image is immediately
uploaded to Google Drive. After successful upload of the file, To make the system wholesome, it was essential to have
the image URL or link to the image corresponding to the further interaction between the user and the system. For the
motion is extracted. same, a web application was developed that allows authorized
users to view information limited to just their surveillance
system.

Fig. 7. Google drive with uploaded images

The URL of the image in the drive for the current motion Fig. 9. Application home page
is extracted and is sent along with start time of the motion in
the body of the e-mail. The user receives an email notification From the home page the user has options to view various
on any device which has the user‟s email account as a graphs that have been plotted using data extracted from RDS.
connected account. The graphs and data visualized here are unique to each user
The e-mail received by the user contains time of start of and their system.
motion as well as image captured by camera as shown in Fig.
8.

Fig. 10. Time Duration graph – 1

On clicking any of the “See Graph” buttons, the

application will render different html pages which contain
different types of graphs. It comes with several interactive
operations such as pan, zoom, save, etc. which can be
triggered by choosing one of the many options listed as icons
Fig. 8. E-mail Notification received by User on the edges of the plot.

978-1-5386-2044-1/17/$31.00 ©2017 IEEE 312

 2nd IEEE International Conference on Computational Systems and Information Technology for Sustainable Solutions 2017

I. SUB-PROCESS DURATION ANALYSIS

Wr ite Upload Upload Trigger

Image on image to time to sending
R Pi Drive RDS email

Average 0.0123 4.0641 0.5177 1.7907

Duration
(sec)

Standar d 0.0023 1.6671 0.0561 0.3986

Deviation
(sec)

Table I: Values are rounded off to 4 decimal places

Fig. 11. Time Duration graph – 2

On choosing which graph the user want to see, Flask II. SYSTEM LATENCY ANALYSIS
renders the html template on which Bokeh produces the output
file as a graph. These graphs are interactive and aid the user in
visualizing and analyzing the data. Time it takes for user Total Process
to get notifie d of
detected motion
The system was later tested in a different environment
where multiple entities were causing motion simultaneously.
The system continued to perform accurately by detecting the
Average 5.8731 6.8818
Duration
motion as well as the causes and responding to this change of (sec)
state, instantly and accurately as shown in Fig. 12 below.
Standar d 1.6672 1.5193
Deviation
(sec)

Table II: Values are rounded off to 4 decimal places

The Total process is the duration to execute all the tasks,

namely writing the image to memory, uploading image to
Google Drive, uploading information such as entry time and
image URL to RDS, triggering sending of e-mail and finally
uploading end time of the motion into RDS.

The average latency between detection of motion and

notification to the user has been calculated as approximately
5.9 seconds. This value can be easily reduced to a much
smaller value of 1.8 seconds by to not uploading the captured
image to Google Drive before sending the e-mail. Although,
the system was made with the aim that the user should receive
the notification for a certain motion in a single e-mail rather
than sending multiple notifications, which would be
Fig. 12. System Detects Multiple Motions and entities
impractical. This decision was made because a near 4.1
Internet of Things expects the processes within such second delay would not induce any severe consequences for a
systems to occur in real time. Hence, the system was tested on video surveillance application, plus the user gets to view the
nearly 5 hours of surveillance video directly from the USB captured image by clicking the image URL added in the e-
camera. 500 entries of times taken for various sub-processes mail.
were logged into a CSV file as the script was running. The
durations for execution of each of these processes were Standard deviation of durations of each of these tasks has
averaged to give us an understanding of latency between been calculated as it is a reliable way to measure how the
various processes that can be expected in such a system. response times and latencies of the system vary.

313 978-1-5386-2044-1/17/$31.00 ©2017 IEEE

 2nd IEEE International Conference on Computational Systems and Information Technology for Sustainable Solutions 2017

VI. CONCLUSION VIII. REFERENCES

Any room or section of one‟s home can easily be put [1] Aamir Nizam Ansari \ Mohamed Sedkyl, Neelam Sharma,
under effective surveillance using the developed system. Upon Anurag Tyagil, „An Internet of Things Approach for Motion
the detection of foreign motion, the user is instantly notified Detection using Raspberry Pi‟ in 2015 International
along with the required visuals he/she requires to ascertain Conference on Intelligent Computing and Internet of Things,
whether the motion is that of an infiltrator or harmless, leaving pp. 131-134
the due course of action to the user‟s discretion. By virtue of
the Internet of Things concepts implemented while developing [2] Huu-Quoc Nguyen, Ton Thi Kim Loan, Bui Dinh Mao and
the system, the user can receive minimal-latency updates in Eui-Nam Huh, „Low Cost Real-Time System Monitoring
any location as long as there exists an Internet connection. Using Raspberry Pi‟ in ICUFN 2015, pp. 857-859
Similarly, the user is also able to access the stored data from [3] Virginia Menezes, Vamsikrishna Patchava, M. Surya
any remote location on a computer or mobile device, provided Deekshith Gupta, „Surveillance and Monitoring System Using
the correct login credentials are entered. Raspberry Pi and SimpleCV‟ during 2015 International
Conference on Green Computing and Internet of Things, pp.
1276-1278
VII. FUT URE WO RK
[4] M. Chen, X.L. Xu, „An Improved Method for Motion
Wide applications of Motion Detection Systems have Detection by Image Difference‟ in Information Science and
surfaced itself in the last few years from tracking to object Control Engineering 2012, 7-9 Dec
classification. A system that uses a deep learning model to [5] Shih-Chia Huang, „An Advanced Motion Detection
classify the objects into simple things like “person”, “car”, and Algorithm with Video Quality Analysis for Video
“door” has been successfully tested on our computers that Surveillance Systems‟ in IEEE Transactions on Circuits and
possess much better performance specifications than a Systems for Video Technology, 18 Oct, 2010.
Raspberry Pi. This method had consumed a lot of processing
time leading to a lot of delay, mainly because of lower RAM
and CPU processing capability of the Raspberry Pi (1GB).

In future, the endeavor is to write a more optimized and

less memory hungry algorithm for a smarter detection and
could also possibly waver off some training datasets that
compromises little on performance.

978-1-5386-2044-1/17/$31.00 ©2017 IEEE 314