1. Panic-driven Event Detection from Surveillance
Video Stream without Track and Motion Features
Mahfuzul Haque and Manzur Murshed
2. Presentation Outline
• Introduction
– Area
– Problem
– Objective
• Event Detection
• The Idea
– Why not track or motion features?
• The Proposed Method
• Experimental Results
• Q&A
3. Research Area
Dynamic Scene Understanding
Stage 1
Video Stream
Stage 2
…
Real-time Processing
Event Detection
Action / Activity Recognition
Behaviour Recognition
Behaviour Profiling
Event
Model
Analytics
Intelligent Video Surveillance
Automated Alert
Smart Monitoring
Context-aware Environments
4. The Problem
Dynamic Scene Understanding
Stage 1
Video Stream
Stage 2
…
Real-time Processing
Scene specific tuning
Availability of training data
Large Surveillance Network
Thousands of video feeds
Ad-hoc remote surveillance
Dynamic scene variations
Event
Model
Analytics
How to develop a generic scene
understanding framework that
would reliably work on a wider
range of scenarios?
5. Research Objectives
Dynamic Scene Understanding
Stage 1
Video Stream
Stage 2
…
Real-time Processing
Event
Model
Analytics
A generic scene understanding framework
Developing the building blocks for the essential processing
stages
Scope:
Panic-driven abnormality detection
A fixed set of specific events
6. Event Detection
time
Specific types of events vs. abnormality
An event persists for a certain duration of time
The duration is variable
Event characteristics of the same event
Variable in the same environment How to identify the generic
Variable from one scene to other
characteristics of an event?
8. The Idea
Motion based approaches
Tracking based approaches
Key points detection
Point matching in successive frames
Flow vectors: position, direction, speed
Object detection
Object matching in successive frames
Trajectories: object paths
Common characteristics
Inter-frame association
Context specific information
Event models are not generic
Hu et al. (ICPR 2008)
Proposed generic approach
Object detection
Global frame-level descriptor:
independent of scene characteristics
Xiang et al. (IJCV 2006)
No Inter-frame association
Independent frame-level features =>
temporal features considering speed
and temporal order
16. The Proposed Method
Top five features for four different events
Feature ranking using absolute value criteria of two sample t-test, based on
pooled variance estimate.
17. Experimental Results
Specific Event Detection
•
•
•
•
•
•
•
Four different events: meet, split, runaway, and fight
CAVIAR dataset with labelled frames
80% of the test frames for model training
100 iterations of 10-fold cross validation
Remaining 20% of the test frames for testing
SVM classifier as event models
Separate model for each event
20. Experimental Results
Abnormal Event Detection
•
•
•
•
University of Minnesota crowd dataset (UMN dataset)
The Runaway event model
No additional training or tuning
Three different sites
25. Experimental Results
Performance Comparison
Method
AUC
Our Method
0.89
Pure Optical Flow [1]
0.84
[1] R. Mehran, A. Oyama, and M. Shah, “Abnormal crowd behavior detection using social force model,” in Proc. IEEE
Conference on Computer Vision and Pattern Recognition CVPR 2009, 20–25 June 2009, pp. 935–942.
26. Publication
Mahfuzul Haque and Manzur Murshed, “Panic-driven Event Detection
From Surveillance Video Stream without Track and Motion Features,”
IEEE International Conference on Multimedia & Expo (ICME), 2010.