Research On Crowd Abnormal State Detection Method Under The Complex Environment

A crowd abnormal state detection based on video is a detection method forabnormal events, which analysis intelligently the behavior of group event, and judgewhether it have crowd trampled, fights, riots in the public places with a large population.The traditional video monitoring relies entirely on the analysis and processing ofman-made and its validity, reliability and rapidity is hard to meet the needs. Therefore,an intelligent detection method for crowd abnormal state based on video is veryimportant. Because of the complexity of group events, the crowd occlusion,illumination and noise, the crowd features is difficult to be extracted. The crowdabnormal state detection is still a problem to be solved.In this paper, the detection for crowd event such as the group fights, riots isresearched. A new crowd abnormal state detection method based on the interaction forceand the local binary pattern co-occurrence matrix is proposed. The main research workis as follows:1) For the problem of inaccuracy crowd motion direction and speed information tobe extracted, the high accurate optical flow algorithm is used. It combines the socialforce model to extract the interaction force, and the velocity and direction of movingcrowd particles under the complex environment are extracted effectively. Then, thetender direction Histogram of Interaction Force (HOIF) is adopted to improve theaccuracy of the crowd motion feature description. It reduces the computationalcomplexity and is robust to illumination and noise.2) According to the gray scale invariant features of the local binary model, and itsgood description of image local texture features, the local binary pattern co-occurrencematrix algorithm are computed respectively on the gray-scale and gradient image toestimate the crowd density, which makes full use of the spatial statistical information ingray image and the edge feature information in gradient image. Then, the energy,contrast, entropy, correlation of the co-occurrence matrix is computed to characterizethe crowd density information. This method improves effectively crowd density featuredescription.3) A new crowd abnormal state detection method of combining the crowd densityinformation with the crowd motion information is proposed to detect the crowdabnormal events, such as fights, riots, trampled. It detects crowd abnormal behavior combining crowd movement behavior with the flow of people, which can describe thecrowd event features effectively. The experimental results show that the proposedalgorithm improves the accuracy of the crowd abnormal state detection.