Abnormal Crowd Behavior Detection Based On Intersection Density Of Motion Vector

Since the 21st century,with the development of social economy,the growth of population,and the occurrence of different kinds of public security incidents,the monitoring and management of crowd behavior in public places have became a hot issue in intelligent video surveillance.However,because of the cover between people,the change of the scene and the unpredictable attitude of the pedestrians,the abnormal human behavior recognition and judgment become a difficult problem in the intelligent monitoring.Aim at the problems above,a new method for the detection of abnormal crowd behavior based on intersection density of motion vector is proposed.When the panic crowd behavior occurs,the crowd instinctively escape all around,which make the motion vector of the crowd different from the normal time.The differences of the motion vector will lead to its intersection becoming more concentrated.So we can make use of these differences to detect the panic behavior of the crowd.The system of abnormal crowd behavior detection mainly includes image denoising,moving information detection and the classification of the crowd.In the first part,a kind of speeded-up K-SVD image denoising algorithm which combines the advantages of the efficiency of median filter and the effect of K-SVD is proposed.In the second part,we use the method of dynamic interframe spacing updating to reduce the algorithmic complexity of pyramid LK optical flow,which not only could detect the motion information of the crowd effectively,but also improve the efficiency of the algorithm.Finally,the SVM classifier is used to classify the behavior of the crowd in video.In this paper,experiments are carried out on the UMN public video set,and the experimental results show that our method can accurately,effectively and quickly identify the abnormal behavior of the crowd.