Research On The Detection Method Of Abnormal Crowd Incidents

The crowd abnormal state detection based on video can be helpful for the public security early warning and the gathered crowd counseling. If the existing resources of the Surveillance system can be applied effectively and intelligently to analysis the crowd behavioral state, and to determine whether there are crowd abnormal events such as stampede, fights, riots, and so on, the disaster events can be avoided.In this thesis, a crowd abnormal state detection algorithm based on the crowd density and velocity characteristics is proposed. It includes crowd motion detection, crowd density detection, crowd abnormal state detection and so on. The main research works are as follows:1) For the crowd density feature extraction, an improved background difference method to be used to extract the foreground image. The texture analysis based on gray level co-occurrence matrix is used to extract energy characteristic value to describe the crowd density in the foreground image. The energy characteristic values of two adjacent frames are applied to determine the change of crowd density. It can decrease the effect of the background.2) For the motion feature extraction, high accurate optical flow algorithm based on deformation theory is used to extract the crowd’s speed, which improves the accuracy of crowd motion extraction. The speeds of two adjacent frames are extracted to obtain the change information of the crowed speed.3) When abnormal events, such as riot, trample, gunslinging, occur, the crowd’s density and motion speed will change. Therefore, the change information of crowd’s density and speed can be combined to detect the crowed abnormal events. In this paper, the method for combining characteristics of the density with the speed is proposed to detect the crowed abnormal events. The experimental results show that the proposed method can effectively detect the occurrence of the crowed abnormal state. It improves the accuracy for crowed abnormal state.