Abnormal Crowd Behavior Detection Based On Wavelet Transform

With the rapid development of economy and society in our country, the urban population density is also growing and great accidents happen easily in crowded areas, causing casualties and property losses. Faced with these increasingly prominent group outburst public events, intelligent monitoring provides an effective information channel to solve this problem. If abnormal crowd incidents can be detected and reported intelligently, not only can relevant personnel be freed from fatigue work of monitoring, and solve the shortcomings of traditional video surveillance alarm delay.For abnormal crowd behavior detection, researchers have proposed a number of relevant models and algorithms, and made some achievements at present. But the existing crowd abnormal detection has lower accuracy and poor real-time performance, and can not meet the actual needs of the application. In crowded public areas, especially in some large-scale markets or subway stations, crowd abnormal analysis faces many problems, like the diversity of the scene, complex interactions of the crowd, mutual occlusion and so on. Therefore, it is necessary to conduct more in-depth research about crowd behavior detection to improve accuracy and time efficiency.For analyzing crowd behavior in surveillance video, we propose a novel method to detect abnormal behaviors in crowd video using Wavelet Transform, which can not only detect dense crowd abnormal behavior but also individual abnormal behavior in sparse population. This method first adopts KLT algorithm to analyze motion information involving in the video sequences. Based on this, crowd collectiveness is defined to measure the consistency of crowd behaviors. Second, Wavelet Transform for presenting unique time-frequency localization is used to capture the singularities in consistency indexes which denote the abnormal behaviors in crowd videos.To evaluate our method, two publicly available datasets-crowed video dataset[39] and UMN dataset are utilized to evaluate our method in detecting abnormal behaviors of dense crowd and sparse crowd. And this method is compared with other detection methods. Experimental results show that the proposed method can capture the dynamics of the crowd behaviors and detect the unusual events in real time.