| The occurrence of mass surveillance video fights,robbery and other bad behavior,artificial eyes to video observation of long time continuous,real-time to the abnormal behavior in video surveillance for detection and identification is the continuous research topic in the field of security intelligence.In this paper,we study the abnormal behavior of the moving body in the complex background environment and the multi-person flow environment.The research content is divided into two parts:the effective tracking and the abnormal behavior recognition of the target.Firstly,the background difference method of the mixed Gaussian background modeling Rapid detection,real-time record of the trajectory of each target,when the target walking normal use of empirical distance with the block histogram matching method to track;when the target occlusion,intersection and loss of the situation,the use of Kalman filter combined with regional light The method of tracking the corner point is used to predict the position of the target motion,and the position tracking is carried out by using the microscopic tracking of the optical point to realize the effective tracking of each target.Second,through the region-weighted optical flow histogram to determine the target may be fighting,and then use the SVM classifier combined with HOG feature training upper body attack behavior classifier to detect whether the fighting action,and finally through the weighted gradient histogram on the detection of the punch Position change histogram change rate,the change frequency is higher than the threshold and sustained more than 20 frames is identified as the fighting behavior;by setting the abnormal speed model,compared to the regional optical flow location information,combined with Kalman filter algorithm for robbery behavior identification.Finally,through several experiments,the above method can effectively identify the two abnormal behaviors of fighting and robbery,and the accuracy rate is higher. |
PDF Full Text Request