Research And Realization Of Detection Method Of Abnormal Crowd Events Based On Population Groups

With the development of computer vision technology and intelligent analysis techniques,research on intelligent video analysis are also constantly improve and enhance,the crowd behavior analysis and research in the video has become a hot research direction.Analysis and study of crowd behavior can be applied in a variety of scenarios,for example:monitoring and early warning to the population in accident in intelligent video surveillance;estimation of the population flow and crowd density estimation large-scale human activities,reduce congestion and prevent a stampede of time.Intelligent video surveillance system is developed in the computer vision technology and intelligent analysis based on the technology of information extraction,can be more effective in the video,and automatic processing of such information,and greatly improve the efficiency of the video monitoring system,to avoid the traditional video surveillance system,monitoring personnel in long time staring at the monitor screen wall,decreased attention missed abnormalities,to reduce the workload of monitoring personnel.On the basis of analysing existing processing methods of intelligent video surveillance,using the processing method of oriented groups features and pre-processing the samples of the video images obtains the features that can represent the whole image,and then detects the abnormal behaviors using the corresponding model.In this paper,firstly,the crowd optical flow feature extraction and crowd physical model are summarized,which achieves Lucas-Kanade and block matching method in the optical flow motion feature extraction,and then compares with the experiments;Next,analysing characteristics of social force model and the rotating social force model,use the rotating social force model to represent motion between the crowd,which designes and achieves extraction method of Size-adapted spatio-temporal cuboid;Finally,proposed a combination of a crowd detection based on Size-adapted spatio-temporal cuboid and rotating the social force model:This method uses block matching method or Lucas-Kanade method to extract features of the crowd motion,which estimates rotating social force with the rotating social force model and detects the event in the abnormal population using LDA model based on the extraction of Size-adapted spatio-temporal cuboid.In the experiment,the normal crowd videos are used to train the model,and the model is used to detect video abnormal behavior.The experimental results show that the method can effectively detect the abnormal behavior of the crowd and verifies the feasibility of the method.