| Human behavior recognition is one of the most important techniques for human motionanalysis. And, it has been widely used in many fields, such as intelligent surveillance, intelligenthome care system, medical gait analysis, athlete support training system, virtual reality and so on.Based on human behavior recognition, abnormal human behavior recognition combines theprocess of object detection, feature extraction, motion modeling and abnormal behavior detectionto analyze the abnormal behavior. Intelligent surveillance with abnormal behavior recognition isnot only to save huge human labors, but also to improve efficiency compared to traditionalsurveillance. A novel abnormal behavior detection algorithm is proposed in this study based onthe analysis and conclusion of the methods available.A method of key-frame extraction based on video contents is proposed to improve theefficiency of the abnormal human behavior recognition. Firstly, a Hu moment method isintroduced to descript the video contents. Then, the coverage rate of the adjacent frames arecalculated to choose the candidate key-frames in a video. Finally, the key-frames are extractedfrom the candidate key-frames using distortion rate as the compression of a video.We propose a human behavior description method based on the multi-features fusion forbehavior modeling. Multi-features including six-star skeleton, angles of six sticks andeccentricity are employed to construct various models of human behaviors. Then,multi-categories supported vector machine with radial basis function kernel is used to constructlearning model for behavior recognition. And, a feature library for kinds of behaviors isestablished from the collection and public videos to generate the multi-categories classifier.11kind of behaviors and the abnormal behaviors in specified scene can be detected withthis algorithm. The precision and time of recognition in experiments with this algorithm issatisfied, which is shown in the end of this paper. |
PDF Full Text Request