You are on page 1of 26

PEOPLE COUNTING & OBJECT

TRACKING
CHIRAG PAI K (1BG11CS026)
KEVIN PRASANNA R R (1BG11CS043)
UNDER THE GUIDANCE OF

DR.B.G.PRASAD

ABSTRACT

Project deals with counting people by finding


contours.

Track object using homography technique.

A comparison between proposed system and


existing machine learning technique.

INTRODUCTION

Need for surveillance system.

Existing Machine Learning techniques & its


complexity.

Existing object tracking techniques.

COMPARISON

PROBLEM STATEMENT

To develop a stable system for :


Effective counting of people at distinct
locations.
Trigger warning when human detected at
restricted area or time.
Trigger warning for missing object when not
found.

LITERATURE SURVEY

Human detection & tracking based on stereo vision (Songmin


Jia, Liang Zhao)
HOG & Kalman filter for video-based human detection &
tracking (Changyan Li, Lijun Guo)
Human detection & tracking using face detection

EXISTING MACHINE LEARNING


TECHNIQUES

Siemens perimeter
tracking system.

Centre of Excellence AntiTerrorist Advanced


Systems

REQUIREMENTS

Software Requirements :
OS : Ubuntu 14.04 LTS (Open Source)
Library : OpenCV 2.4.9 (Open Source)

REQUIREMENTS

Hardware Requirements :
CPU : Clocked over 1.6 GHz
RAM : 2GB
Camera / Video Footage

SYSTEM DESIGN
CONTEXT
DIAGRAM

LEVEL-0 DFD

LEVEL-1 DFD : OBJECT


TRACKING

LEVEL-1 DFD : PEOPLE


COUNTING

FLOWCHART

PEOPLE
COUNTING

OBJECT
TRACKING

TESTING
Test ID

Test Case

Expected
Output

HOG on Main
Bldg

5 People

20 People

HOG on New
Bldg

5 People

24 People

Proposed
Model Main
Bldg

5 People

5 People (+1,-1)

Proposed
Model New
Bldg

6 People

6 People

Actual Output

TESTING
Test ID

Test Case

Expected
Output

Actual Output

Book Image

Object
Detected

Object
Detected

Book Video

Object
Detected

Fluctuation

Lab CRO Image

Object
Detected

Object detected

RESULTS

CONCLUSION

The developed system is stable.

Robust to lighting changes.

Object tracking is highly accurate.

People counting is accurate in most cases.

LIMITATIONS

Occlusions are not dealt with.

Shadow intervenes with people count.

Need for templates for object tracking.

FUTURE ENHANCEMENTS

Ability to track individuals using machine learning.

Usage of latest background subtraction algorithm


& filters.

Increase robustness of people counting in indoor


environments.

REFERENCES

Background Subtraction & Segmentation:

Z.Zivkovic, Improved adaptive Gausian mixture model for background subtraction, International Conference Pattern Recognition, UK, August,
2004, http://www.zoranz.net/Publications/zivkovic2004ICPR.pdf

HOG - Histogram of Oriented Gradients &


SVM - Support Vector Machines:

Human Detection by Quadratic Classification on Subspace of Extended Histogram of Gradients,Image Processing, IEEE
Transactions on (Volume:23 , Issue: 1 ),22 May 2013.

Feature Matching + Homography:

Accurate Object Tracking Based on Homography Matrix,International Conference on Computer Science & Service System (CSSS), 2012

You might also like