Trends in Computer Vision Automatic Video Surveillance. Overview Why do we need automatic surveillance for Criminal and Anti-social behaviour detection? Research Issues and some solutions Commercial Solutions? Ethical and Moral Issues Conclusions and Future Developments Why do we need Automatic Surveillance? • A surveillance control room operator monitors up to 50 cameras simultaneously. • More and more cameras are being placed in public areas. • Recognition/prediction of violent/antisocial or criminal acts. Researched Solutions • Potential anti-social and criminal behaviour between people can be predicted by humans. – There are key types of body motion that allow predictions to be made. – Will They Have A Fight? The Predictability of Natural Behaviour Viewed Through CCTV Cameras, Troscianko et al. European Conference on Visual Perception 2001 Researched Issues and Some Solutions • Building blocks for automatic surveillance. Extract features Video Scene Track people/vehicles Learn Topology of Scene Detect Unusual Behaviour Action/Gesture Recognition Researched Issues and Some Solutions • Feature Extraction – sdfsdfs Extract features Video Scene Track people/vehicles Learn Topology of Scene Detect Unusual Behaviour Action/Gesture Recognition Researched Issues and Some Solutions • Learning Scene Topology – Occlusion Analysis: Learning and Utilising Depth Maps in Object Tracking Greenhill et al, British Machine Vision Conference 2004 – Learning Spatial Context from tracking using Penalised Likelihoods, McKenna and Nait-Charif, International Conference on Pattern Recognition 2004 Extract features Video Scene Track people/vehicles Learn Topology of Scene Detect Unusual Behaviour Action/Gesture Recognition Researched Issues and Some Solutions • Tracking people/vehicles – Tracking Multiple Humans in Crowded Environment, Zhao and Nevatia, Conference on Computer Vision and Pattern Recognition 2004 – Rapid Object Detection using a Boosted Cascade of Simple Features, Viola and Jones, Conference on Computer Vision and Pattern Recognition 2001 Extract features Video Scene Track people/vehicles Learn Topology of Scene Detect Unusual Behaviour Action/Gesture Recognition Researched Issues and Some Solutions • Unusual Activity Detection – Detecting Unusual Activity in Video, Zhong et al, Conference on Computer Vision and Pattern Recognition 2004 Extract features Video Scene Track people/vehicles Learn Topology of Scene Detect Unusual Behaviour Action/Gesture Recognition Researched Issues and Some Solutions • Action/Gesture Recognition Extract features Video Scene Track people/vehicles Learn Topology of Scene Detect Unusual Behaviour Action/Gesture Recognition Commercial Solutions • Most software is based around motion sensors. Very few deal with real-time intelligent video processing. • OpenCV – provides some open source tools for making your own commercial systems ( for a fee!). E.g face detection Commercial Solutions • Safehouse Technology Ltd – demo on face detection – demo on appartment block Ethical and Moral Issues • Advantages: – Potential for quicker response times. – Frees up law enforcement resources to chase other more complex crimes. • Disadvantages: – Big Brother • We are being captured by more and more security cameras. – Face Recognition • Concerns about people having access to large databases of faces. Conclusions and Future Developments • There is still a long way to go… • Integrated camera systems for cross-camera criminal event detection.