Research On Abnormal Behavior Detection Based On Video Surveillance

The detection of abnormal behavior of pedestrians has gradually become a popular topic in the field of intelligent video surveillance.Pedestrian safety is not only related to pedestrians but also affect the surrounding traffic system for the complex indoor and outdoor traffic environment,.However,the amount of surveillance data and Internet data has springed in recent years.Previous safety management schemes relay on artificial can not satisfied the era of big data.At present,there is a great demand for intelligent video surveillance system with efficient and reliable,and the intelligentize of video monitoring system needs to be improved.In this paper,weak supervised detection method of abnormal behavior based on temporal consistency is proposed.First,temporal gram matrix are constructed for a given pair of video.Then a pair of behavior units(Candidate action fragment)is formed by exploiting the temporal consistency and smoothness of human behavior.The target is to locate the start and end frames of the related abnormal action class in the video,and training the corresponding classifier.At last,the local feature of histogram of sparse code(HSC)and the method of sparse reconstruction are used to accurately locate the abnormal events in the monitored video.Before the detection of abnormal behavior,this paper performs the minimum spanning tree and distance transformation to the superpixels in the image.Then fuzzy color difference histogram processing was carried out to the image background,which effectively reduces the false detection probability.In the weakly supervised detection method,every action separates the same number of subactions,and each video is positioned according to these sub behaviors.After getting the complete frame,the local feature of sparse coding histogram and the method of sparse reconstruction are used to accurately locate the abnormal events in the monitored video.At last,the double threshold method are used to discriminate abnormal behaviors,which improves the recognition accuracy.The effectiveness and robustness of this method is verified in the public database of Crossing and Subway,which also compared with other popular methods.The results show that the method still can carry out the detection in the presence of light and shadow interference situation.Our method have certain improvement in performance compared to other methods.