Research On Crowd Abnormal Detection Method Based On Video

Crowd abnormal state detection based on video is a detection method which refers to analyze state of behavior of mass events of large crowds in public place, and detect whether there is a crowd of abnormal event. In this thesis, detecting crowd abnormal events are focused on, and a detection method is adopted, which combines the crowd motion information with the crowd density information. The main research tasks are as follows:1) For the complex computation of extracting the crowd motion information, crowd motion vector is obtained by high accuracy optical flow firstly. Then, motion direction distribution histogram is used to get crowd motion direction features, and motion vector intensity distribution is adopted to extract motion intensity features. The motion intensity features with the motion direction distribution histogram are combined to describe the crowd motion information, which reduces the computational complexity effectively and guarantees accuracy of crowd movement state features description.2) In order to use the gray-scale spatial statistical information and edge feature information of video frames effectively, in this thesis, the co-occurrence matrices are acquired based on local binary pattern of the gray image and gradient image for video frames separately. The contrast, energy, entropy and correlation are used to describe the population density information, which improves the effectiveness of the crowd density information.3) In the thesis, combines the crowd motion features with the population density features are combined, and support vector machine is used to detect crowd abnormal events. The experimental results show that the proposed algorithm effectively reduces the computational complexity and guarantees the accuracy of crowd abnormal detection.