Skip to Main Content
The paper investigates a novel and effective approach for real-time analysis of crowd congestion (density) in a physical space monitored by surveillance cameras. A region of interest (ROI) is specified in the space and partition of the ROI into an irregular array of sub-regions (blobs) automatically carried out, to each of which a congestion contributor is computed. The method then exploits the short-term responsive background (STRB) model for blob-based dynamic congestion detection, and uses the long-term stationary background (LTSB) model for blob-based 'zero-motion' (static congestion) detection. A global feature analysis is adopted for scene scatters characterisation; and finally, the combination of the local and global analysis gives the accurate scene congestion rating. Besides, this scheme is adapted to perform the task of moving object presence detection with success. Extensive tests and field trials validate both the accuracy and robustness of the approach.