Application And Research Of The Pedestrianabnormal Behavior Detection Technology Based On The Video Image

The intelligent video surveillance is a kind of technology which uses computers to replace manual operations to analyze the contents of the surveillance video. This technology has been more and more popular in the area of defense and security. Generally analyzingthe behavior of the pedestrian in the video is a main subject in surveillance. So detecting abnormal behavior of human in the video is the key issue for intelligent video surveillance. In this paper, we focus on doing basic research about detecting abnormal behavior of the pedestrians in the video,meanwhile, the following proposed algorithm and methods have been applied to the practice application. The thesis proposed in this paper probes into detecting anomalies of people with the videos as input.In this paper, we divide the behavior of people in the video into two kinds. One is the single abnormal behavior and the other is the anomalies of the crowd. First of all, we detect and track the single walker in the video on single analysis. Then we make a comparison of the outline we extracted and the outline in the standard template to judge the behavior. But when we analyze the crowd, we extract a topology from the crowd macroscopically. We find that, if the topology of changes there must be some varieties happened to the crowd. So we can use the computer to detect the anomalies replaced the human monitoring when the topology changes.We adopt a lot of optimizations in the two approaches we proposed. In the single analysis, we use a new detecting model with slow and fast updating policies. The method of combination of Kalman filter and color model has been applied to detecting the moving foreground. When we extract the contour we compensate for it and reduce the dimension of it. The minimal similar degree template generation and the optimized Hausdorff distance which used to measure the image similarity are presented, which can solve the problem brought up by the uncertain speed motion. We apply the topological simplification algorithm to analyzing the behavior of thecrowd, meanwhile, we extract the topology which can stand for the overall motion of the crowd more precise. We monitor the change of the topology to detect the varieties of the behavior of the crowd, such as formation/dispersion and splitting/merging. This approach is simpler and more stable compared with the classical method which need a complex classifier constructed.