Research On Detection Of Human Abnormal Behavior In Video Sequences

Video processing technology is widely applied in national economy and national defense as development of copmputer and information processing techniques. Huge video data produced in many applications needed to be processed, especially in video surveillance. Then video image analysis and processing technology is becoming an important direction in research of information processing in oorder to improve its intelligent degree.The thesis focuses on the human abnormal behavior detection, which includes moving human detection, tracking and behavior analysis, detection of abnormal behavior.After introducing relative basic theories, existing approaches of moving human detection in static background are reviewed in the thesis, and a new processing procedure is developed. It consists of three steps: background modelling with Gaussian mixture model, foreground detection with background substraction and its following processing.A new detection and tracking approach of a moving person is developed in this thesis, which is an integrated approach with particle filter and level set. Its procedure is as following: the moving region with rectangle shape including the moving person is detected using particle filter, the contour of the human figure is extracted by level set method, and the contour result extracted is feedback into tracking scheme in order to improve likelihood function by dynamically updating the reference template. The approach has adaptability to arbitrary variation of person moving, and high tracking precision.An abnormal behavior detection approach of moving person is addressed in the thesis. It concludes: moving person is tracked with region features in video sequences, the parameter curve tracked is formed and fitted, and abnormal behavior of a person is distinguished based on the fitted curve. It is simple, fast and high efficient.