Research on the Crowd Abnormal Behavior Recognition in Surveillance Video Based on Modified Social Force Model
Meng Kang, Lei Xia, Hao Chen · 2019
A surveillance monitoring public places is an effective means to prevent security problems, but traditional surveillance can't make rapid respond to existing security problems and cannot satisfy social requirements. For that matter, this paper proposes a crowd abnormal detection algorithm based on modified social force model (MSF)to recognize the abnormal behavior of panics and flees caused by emergencies in public. Based on the conventional social force model (CSF), this paper combines the research results of predecessors to improve the expected speed, in the consideration of the influence of relative speed between pedestrians at different locations in the social force calculation, it adopts the positional state factor and the relative speed influence coefficient to improve the social force model. This paper uses Lucas-Kanande algorithm to extract image optical flow, clusters optical flow information, and then uses modified social force model to extract the crowd motion features, and finally applies support vector machine (SVM) to recognize abnormities. In the UMN dataset test, the area under ROC curve is 0.9963.