Abnormal Crowd Behavior Detection and Location Based on Spatial-temporal Cube

To handle the issue of low accuracy performance of crowd abnormal behavior detection in video surveillance systems, an abnormal crowd behavior detection and location approach based on spatial-temporal cube is proposed in this paper. The optical flow method is first used to calculate the optical flow field of feature points which are obtained by the equidistant sampling method. Then, the velocity, orientation and orientation entropy of the feature points are obtained. And statistical histograms of the three parameters are mapped into the corresponding cubic space to extract the spatial-temporal cube feature for describing the spatial-temporal features. A blocking method is used to divide the image into several sub regions, and the spatial-temporal cubes of each sub region are calculated. Finally, a cascade classifier based on nearest neighbor classification and support vector machine is designed to detect and locate crowd abnormal behaviors. Experimental results show that the proposed method can effectively detect and locate abnormal crowd behaviors in videos.