Anomaly Detection in Surveillance

Learning Abnormal Behavioural

for Surveillance Systems

By Ahmed Ibrahim

23/2/2011

Outline

• Introduction

• Key Challenges

• Proposed approaches

• Preliminary experiments

• Discussion and future directions

What is Abnormal Behaviour ?

• Abnormal Behaviour is a pattern in the data

that does not conform to the expected normal

behaviour.

• Also referred to as outliers, exceptions,

suspicion, surprise, etc.

• Example: set of data points

Feature space.

F1

F2

N1

N2

o1

o2

O3

Video Surveillance Example

A car on pedestrian roadway

A piece of luggage left in a check in-area

Source: Performance Evaluation of Tracking and Surveillance dataset (PETS)

Key Challenges

• Defining a representative normal behaviour is challenging.

• The boundary between normal and outlying behaviour is often not precise.

• The exact notion of an outlier is different for different video surveillance applications.

• Availability of labelled data for training/validation.

• Data always contain noise.

• Normal behaviour keeps evolving

Abnormal Behavior Detection

Framework

Motion Detection

• Background

subtraction

• Temporal

differencing

• Optical flow

Object Tracking

• Model

• Region

• Active contour

• Feature based

Behavior

Understanding

• Classification

(supervised)

• Clustering

(unsupervised)

Behavior Type

• Label

• Score

Source: PETS

Unsupervised Behavior Modeling

•The following trajectory

has been generating by:

•Applying principal

component analysis

on the video stream;

•Selecting the first

three components

•Every point on the

trajectory represents a

frame from the video

stream

Learning Outliers

Behavior

Model

Gathering

Real Data

Statistically

Resampling

Visually

Resampling

Proposed Statistical Approach

Statistical Resampling Example

Subspace

trajectory of

waking pedestrian

in outdoor

PC: for principal components

Irregular segments

Proposed Visual Approach

Visual Resampling Example

Subspace

trajectory of

waking pedestrian

Subspace

trajectory of

synthetic

pedestrian

Time delay

Similar segments PC: for principal components

Behavior Model Output

• Label : each test instance is given a normal or

anomaly label

• Score: each test instance is assigned an

anomaly score

• Allows the output to be ranked

• Requires an additional threshold parameter

Research Plan

• To test the feasibility of statistical resampling: – The Ionosphere dataset from UCI machine learning

Repository will be used.

– This dataset are radar signals sent into the ionosphere and the class value indicates whether or not the signal returned information “Good” or “Bad” on the structure of the ionosphere.

• To test the feasibility of visual resampling: A set of animated videos with real backgrounds will be generated for the following events:– Walk, Run, Jump, Gallop sideways, Bend , One-hand wave,

Two-hands wave, Jump in place, Jumping Jack, Skip.

References

• [1] Weiming Hu; Tieniu Tan; Liang Wang; S. Maybank; , "A survey on visual surveillance of object motion and behaviors,"Systems, Man, and Cybernetics, Part C: Applications and Reviews, IEEE Transactions, 2004

• [2] Monekosso, Remagnino, ”Synthetic Training Data Generation for Activity Monitoring and Behavior Analysis” , Proceedings of the European Conference on Ambient Intelligence, 2009.

• [3] Garg, Aggarwal, Sofat, ” Vision Based Hand Gesture Recognition”, 2009.

• [4] Poppe, “A survey on vision-based human action recognition”, Image and Vision Computing Journal, 2010.

• [5] Mitra, Acharya, ”Gesture Recognition: A Survey”, IEEE transactions on systems, man, and cybernetics, 2007.

• [6] Wang, Suter, “Recognizing Human Activities from Silhouettes: Motion Subspace and Factorial Discriminative Graphical Model”, IEEE Conference on In Computer Vision and Pattern Recognition, 2007.

• [7] Hua, “Probabilistic Variational Methods for Vision based Complex Motion Analysis”, PhD dissertation, Electrical and Computer Engineering, 2006.

• [8] Incertis, Garcia-Bermejo, Casanova, "Hand Gesture Recognition for Deaf People Interfacing", 18th International Conference on Pattern Recognition, 2006.