Student: Dane Brown 2713985 Supervisor : James Connan and Mehrdad Ghaziasgar.

SUSPICIOUS ACTIVITY DETECTION

Student: Dane Brown 2713985

Supervisor : James Connan and Mehrdad Ghaziasgar

OVERVIEW

INTRODUCTION

DESIGN DECISIONS

IMPLEMENTATION

PROJECT PLAN

DEMO

INTRODUCTION Extremely high crime rate in South Africa

Car break-in rate was 16000 in 200918 times the rate of USACarjacking is the most common crime in South AfricaCosting tax payers billions of rands!

INTRODUCTION cont.

2006 2007 2008 200913000

13500

14000

14500

15000

15500

16000

16500

Carjackings 2006-2009

INTRODUCTION cont. CCTV cameras

Human monitoredCurrent solution ineffectiveContinued high break-in rate

INTRODUCTION cont.

Pioneered revolutionary system

Uses computer vision techniquesAutomatically detects suspicious activity from a

video feedDetection happens in real-time

INTRODUCTION cont.

Pioneered revolutionary system

DESIGN DECISIONS Classification methods

Machine learning such as Haar-like features with Adaboost

Generally training 2000+ sample frames

Why not a classification method?

Trade-off between speed, complexity and accuracy

There are simpler and more robust ways to

differentiate suspicious and normal behaviour.

IMPLEMENTATION

Original frame in RGB colour

IMPLEMENTATION cont.

Gray Scale and Frame differencing

IMPLEMENTATION cont.

Motion History Image (MHI)

IMPLEMENTATION cont.

Blob and movement detection (using MHI)

IMPLEMENTATION cont.

Blob and movement detection

IMPLEMENTATION cont.

Blob and movement detection

IMPLEMENTATION cont. System determines normal activity

Park car

IMPLEMENTATION cont. System determines normal activity

Park car

IMPLEMENTATION cont. System determines normal activity

Get out

IMPLEMENTATION cont. System determines normal activity

Walk away

IMPLEMENTATION cont. System determines normal activity

Walk away

IMPLEMENTATION cont. System determines normal activity

Get back in

IMPLEMENTATION cont. System determines normal activity

Drive away

IMPLEMENTATION cont. System determines normal activity

Drive away

IMPLEMENTATION cont. System determines suspicious activity

Loitering next to a vehicle is suspicious

IMPLEMENTATION cont. System determines suspicious activity

Loitering next to a vehicle is suspicious

IMPLEMENTATION cont. System determines suspicious activity

Loitering next to a vehicle is suspicious

IMPLEMENTATION cont. System determines suspicious activity

Loitering next to a vehicle is suspicious

IMPLEMENTATION cont. System determines suspicious activity

Loitering next to a vehicle is suspicious

IMPLEMENTATION cont. System determines suspicious activity

Loitering next to a vehicle is suspicious

IMPLEMENTATION cont. System determines other suspicious activity

Parking, but not leaving the vehicle

IMPLEMENTATION cont. System determines other suspicious activity

Accelerating too fast

IMPLEMENTATION cont.

Suspicious activity detected!

1. Normal activity - typical drive away

2. Suspicious - two men loitering

3. Suspicious - Stationary

4. Suspicious - Acceleration

DEMO

REFERENCES

Davis, J. W. (2005). Motion History Image. Retrieved 2010, from The Ohia State University.

Green, B. (2002). Histogram, Thresholding and Image Centroid Tutorial. Retrieved 2010, from Drexel University site.

Trip Atlas. (2010). Retrieved from Carjacking: http://tripatlas.com/Carjacking#South%20Africa

  Hijacking. (2010). Retrieved from Arrive Alive:

http://www.arrivealive.co.za/pages.aspx?i=2364 

QUESTIONS AND ANSWERS

Thank You!