Research On Human Abnormal Activity Recognition In Video Surveillance

Human abnormal behavior recognition denotes an automatic analysis of humanbehavior in video sequences, and determines whether a human movement behavior isanomaly, which is the research hotspot of computer vision field. The research on thehuman abnormal action recognition technology plays an important role in intelligentvideo surveillance system. It has great practical significance. A system based onabnormal behavior analysis can not only remove the useless information, but also fulfillthe task efficiently, save large amount of human power, financial resources as well asmaterials, which brings enormous economic benefits to society. In this paper, therecognition method for seven abnormal human behaviors, such as running, walking withwaist bent, jumping, crouching, tumbling, wandering and hitting cars, is researched,which content involves the human movement detection, tracking, feature extraction andrecognition technology.In this paper, several common moving object detection methods, such as framedifference, optical flow and background subtraction, are analyzed and compared.Mixture Gaussian model based on the K-means clustering is selected to detect motionobjects. Then, mathematical morphology is used to conduct subsequent processing toget complete moving objects. Kalman Filtering algorithm is combined with mean shiftalgorithm to improve the real-time performance of moving object tracking,LBP operator and Hu moment are studied in this paper. The features based on LBPoperator can fully describe the image texture feature and Hu moment can wellcharacteristics of moving region information. Therefore, a moving feature descriptionmethod is proposed, which combines LBP features based on uniform rotation invariancemodel and four low order number Hu momentsM1, M2,M3,M4, and has a wellrobustness. The experimental results show that the moving feature description methodcan characterize human motion behavior effectively.For behavior recognition, the method based on the state diagram switching modelis applied. Because Hidden Markov model has good robustness to the tiny changes ofthe temporal and spatial scales movement, It is used to recognize human abnormalbehavior. The experimental results show that the Hidden Markov model for humanbehavior identification can obtain higher recognition rate.