Uses OpenCV and NumPy modules. Detects unusual behavior via the available dataset. Machine Learning Concept.

1. AN INDUSTRY ORIENTED
MAJOR PROJECT
ON
“SUSPICIOUS ACTIVITY DETECTION”
Computer Science & Engineering
Mahaveer Institute of Science & Technology
2019-2020

2. Our Team
Programmer
Working on creating and developing
modules of the software.
Module Designing
Dr. R. NAKEERAN
HOD. CSE Dept
Under the Guidance of
K.SUDHAKAR
Asst. Professor/Assoc. Professor
GAJULA ANJALI PRABHU SANKEERTH
[16E31A05B9]
MUHAMMAD MUSHAHID ALI
[16E31A05D2]
Documentation
Made available of all the documents,
policies and terms.
Specify Documents/Policy
Project Coordinator
SRINIVAS REDDY
Asst. Professor/Assoc. Professor

3. Agenda
Topics
 Abstract
 Introduction
 Existing System
 Proposed System
 Hardware Requirements
 Software Requirements
 System Modules
 Implementation Output Screens
 Conclusion
 References

4. ABSTRACT
4
With the increase in the number of anti-social activities that
have been taking place, security has been given utmost
importance lately. Many organizations have installed CCTVs
for constant monitoring of people and their interactions.
01
For a developed country with a population of 64 million,
every person is captured by a camera ~ 30 times a day.
02
A lot of video is generated and stored for certain time
duration( India: 30 days). A 704x576 resolution image
recorded at 25fps will generate roughly 20GB per day.
Since constant monitoring of data by humans to judge if
the events are abnormal is a near impossible task as it
requires a workforce and their constant attention. This
creates a need to automate the same. Also, there is a
need to show in which frame and which parts of it contain
the unusual activity which aid the faster judgment of that
unusual activity being abnormal. The method involves
generating motion influence map for frames to represent
the interactions that are captured in a frame.
03

5. 5
Introduction
 In this project we need to detect person behaviour as suspicious
or not, now a day’s everywhere CCTV cameras are installed
which capture videos and store at centralized server and
manually scanning those videos to detect suspicious activity
from human required lots of human efforts and time. To
overcome from such issue author is asking to automate such
process using Machine Learning Algorithms. To automate that
process first we need to build training model using huge number
of images (all possible images which describe features of
suspicious activities) and ‘Convolution Neural Network’ using
TENSOR FLOW Python module. Then we can upload any video
and then application will extract frames from uploaded video and
then that frame will be applied on train model to predict its class
such as ‘suspicious or normal.

6. Existing Proposed
The existing
system is based
on manual
checking, this
requires
manpower and
time and we
would not get any
exact result.
To automate that
process first we need to
build training model
using huge number of
images (all possible
images which describe
features of suspicious
activities) and
‘Convolution Neural
Network’ using TENSOR
FLOW Python module.

7. HARDWARE
REQUIREMENTS
PROCESSOR
Intel Pentium
or Intel i3
RAM
1024MB to
4096MB
(1-4GB)
Monitor
15 inch
color
Hard Disk
500 GB
KeyBoard
Standard
102 Keys

8. SOFTWARE
REQUIREMENTS
OPERATING SYSTEM
Windows10
Technology
Python 3.6
IDE
PyCharm

9. 9
Optical flow of blocks (optFlowofblocks.py)
The module optical flow of blocks is
provided with a frame and the optical flow of
a frame. It divides the frame into blocks of
size m * n and sums all the optical flows in
each block and returns it along with details
like m, n, size and center of blocks.

10. 10
Motion Influence Generator
(motionInfluenceGenerator.py)
This module is provided with
training or testing video and it
calculates the motioninfluence map
for each frame in that video and
also returns the size of the blocks in
the motioninfluence map.

11. 11
Megablock Generator (createMegaBlocks.py)
This module has 2 functionalities :
a) Generating megablocks and returning them (testing)Megablocks are
generated by grouping motion influence blocks into a bigger sized blocksas
motions of closely situated blocks are similar. A set of megablocks of size
(number of frames* number of megablocks in each row * number of
megablocks each column) is returned.
b) Generating megablocks and returning codewords (training)After repeating
the above process but before returning the set of megablocks, each set
ofmegablocks present in the same frame position is applied kmeans
clustering on and the meanscalled codewords are only returned to the
calling module.

12. 12
Training module (training.py)
Training module calls motion
influence generator and megablock
generator to obtaincodewords on a
training video input. It then stores
codewords in a .npy(NumPY file).

13. 13
Testing module (testing.py)
Testing module calls motion influence generator
and megablock generator to obtain megablocks
on a testing video input. It then constructs a
minimum distance matrix after loading the stored
codewords, checks if a megablock is unusual by
comparing it against a threshold value and
displays unusual megablocks and frames.

14. Simple PowerPoint Presentation
Simple PowerPoint Presentation
Simple PowerPoint Presentation
Simple PowerPoint Presentation
You can simply impress your audience and
add a unique zing and appeal to your
Presentations. Easy to change colors, photos
and Text. You can simply impress your
audience and add a unique zing and appeal to
your Presentations.
OUTSCREENSLIDES….
THE
OUTSCREENS
D E S I G N O U T L E T S

15. 15
CONCLUSION
Finally,
Thus the Suspicious Human Activities can
be detected using this system. Further, this
system can be extended to detect and
understand the activities of people in
various scenarios. This system is currently
developed for detecting the activities of
people in a stationary background. This
system can be further extended to detect
human activities in places with mobile
background.

16. REFERENCE
[1] Dong-Gyu Lee, Heung-Il Suk, Sung-Kee Park and Seong-Whan Lee “Motion Influence Map for
Unusual Human Activity Detection and Localization in Crowded Scenes” IEEE transactions on circuits
and systems for video technology, vol. 25, no. 10, October 2015
[2] Data set – http://mha.cs.umn.edu/Movies/Crowd-Activity-All.avi
[3] Data set - http://www.svcl.ucsd.edu/projects/anomaly/
[4] T. Xiang and S. Gong, “Video behavior profiling for anomaly detection,” IEEE Trans. Pattern Anal.
Mach. Intell., vol. 30, no. 5, pp. 893–908, May 2008.
[5] F. Jiang, J. Yuan, S. A. Tsaftaris, and A. K. Katsaggelos, “Anomalous video event detection using
spatiotemporal context,” Comput. Vis. Image Understand., vol. 115, no. 3, pp. 323–333, 2011.
6] B. D. Lucas and T. Kanade, “An iterative image registration technique with an application to stereo
vision,” in Proc. 7th Int. Joint Conf. Artif. Intell., San Francisco, CA, USA, Aug. 1981, pp. 674–679.
[7] OpenCV Python documentation at http://docs.opencv.org/3.0-
beta/doc/py_tutorials/py_tutorials.html
[8] OpenCV references at http://opencv-python-tutroals.readthedocs.io/en/latest/ 16

17. ANY QUERIES?

18. Thank You!