Research On Abnormal Crowd Activity Detection Algorithm In Video Sequences

With the development of society, the requirements for intelligent life have become urgent. For the safety of crowd, we need a better video surveillance. However, the traditional video is non-intelligent, so there are lots of drawbacks in both its performance and economic value. The intelligent video surveillance appears because of its potential economic value and social needs. Therefore it receives attention of the majority of scholars and became a hot research now.Because of the public safety, this article wants to achieve an algorithm which can detect whether the crowd is abnormal or not. There are four abnormal events-demonstrations, riots, sit-ins and onlookers we should detect when it happens. Though the learning and research on the digital image processing, according to characteristics of the four kinds of unusual events, this paper presents a algorithm which can analysis the population characteristics, in order to distinguish the for abnormal events above.This paper introduces the optical flow method to statistic the movement direction as the main feature to judge the population movement. Be different with the traditional detection algorithm for dynamic abnormal event, this paper simultaneously judge both static and dynamic ones. After judging the crowd's status, this paper uses different algorithms to detect the abnormal event, which expands the algorithm's scope of application. When the crowd's status is static, we adopt the advanced multi-target detection algorithm to extract the typical persons as the crowd's representative, detect their action in order to judge whether crowd is abnormal or not.In addition, this algorithm is different with the algorithms which statistics signal characteristic. This paper estimated the crowd's movement, at the same time it estimate the crowd's density by using Gray Level Dependence Matrix(GLCM)algorithm. By combing the crowd's density with movement effectively, we get a standard to judge whether crowd is abnormal or not. It can prove that the algorithm described in this article can effectively work well after lots of tests.