Human Abnormal Behavior Detection In Surveillance Video

With the improvement of living standards and the development of science and technology,the demand for cameras in public places and home environments is getting higher and higher,and traditional camera monitoring technology can no longer meet people’s safety needs.Therefore,designing an intelligent surveillance system that can quickly find useful information from a large number of surveillance videos is one of the future development directions of surveillance systems.As a part of the intelligent monitoring system,the detection of human abnormal behavior has gradually attracted everyone’s attention.This article mainly detects abnormal behavior in the public environment.First of all,this paper studies a video target tracking algorithm combining Mean Shift target tracking and Kalman filter for the problem of tracking failure when the detected target is severely occluded.When the target is severely occluded,the Kalman filter prediction mechanism is used.According to the above The motion information of one frame of target is used to predict the target motion information of the next frame,thereby eliminating the influence of occlusion on target tracking.This algorithm can overcome the tracking failure problem that occurs when the target is severely occluded,and can improve the accuracy and real-time performance of target tracking.Secondly,when performing abnormal behavior detection,extract the shape features of the target as the body changes when the target moves,representing the compactness of the actual proportion of the detected target in the bounding box and the center of mass velocity when the target moves,and respectively These three characteristics are used as the classification criteria for abnormal behavior.When only one of the features is used for detection,the problem of inaccurate detection is prone to occur.Therefore,this paper adopts the method of combining these three features for behavior detection.Compared with single feature detection,the accuracy is higher and the detection result is more accurate.Finally,when detecting the human-vehicle interaction behavior in surveillance video,an enhanced cascade method is first used to detect the vehicle,and then an improved adaptive boosting(Adapitive Boosting,Ada Boost)algorithm is used to integrate multiple HAAR features.Learning,and finally by combining multiple strong classifiers to achieve vehicle detection.After the vehicle is detected,the abnormal behavior is detected by calculating the correlation between the detection target and the vehicle when the interactive behavior is generated,so that the detection result is more accurate and the behavior classification effect is better.