Crowd Anomaly Detection for Automated Video Surveillance

Video-based crowd behaviour detection aims at tackling challenging problems such as automating and identifying changing crowd behaviours under complex real life situations. In this paper, real-time crowd anomaly detection algorithms have been investigated. Based on the spatio-temporal video volume concept, an innovative spatio-temporal texture model has been proposed in this research for its rich crowd pattern characteristics. Through extracting and integrating those crowd textures from surveillance recordings, a redundancy wavelet transformation-based feature space can be deployed for behavioural template matching. Experiment shows that the abnormality appearing in crowd scenes can be identified in a real-time fashion by the devised method. This new approach is envisaged to facilitate a wide spectrum of crowd analysis applications through automating current Closed-Circuit Television (CCTV)-based surveillance systems.