A Weighted MHOF and Sparse Representation Based Crowd Anomaly Detection Algorithm

In the field of computer vision, crowd video analysis especially crowd anomaly detection had received increasing attentions of research. Effective prediction and detection of the abnormal events in a crowded scene is quite crucial to establish a safe and efficient public environment. In this paper, a more effective algorithm for anomaly detection is proposed based on WMHOF(Weighted Multi-Histogram of oriented Optical Flow) in the framework of sparse representation based algorithm. On basis of the MHOF feature, an energy based weight is introduced to increase its ability of group behaviors describing. Experimental results show that the proposed WMHOF feature is more sensitive to the movements in the scene, so as to establish a more effective normal behavior model for detecting of the abnormal ones. By defining of a sparse reconstruction cost function, the AUC (the Area Under the Curve get a 1%~2% improvement compared with other similar methods.