Research Of The Abnormal Behavior Detection Based On The Improved Algorithm Of Codebook Model And Particle Filter

With the rapid development of computer technology, the intelligent monitoring system has been widely used in military, traffic, security and other fields. So how to improve the accuracy of analysis for the the moving targets' abnormal behavior has become the hot topic in intelligent video surveillance. Using computer image processing technology, abnormal behavior analysis system could detect and track the moving target automaticly and then analysis and judge their behavior, identify abnormal behavior and send alerts in time. The system has good real-time performance, low investment and high accuracy. It is important to detect and track accurately before analysising the abnormal behavior.The algorithm of moving object detection and tracking which was applied to the video images with complex background was deeply studied in this thesis. And the moving targets' behavior was analysised in this thesis. After that, the abnormal behavior system was designed. Aimed to increase the accuracy of the detection and track, the Codebook background model algorithm and the Particle filter algorithm were implemented and improved in this thesis. Main work of this thesis includs the following aspects:1. The traditional Codebook algorithm has high computational complexity, bad segmentation result and poor anti-jamming ability because it sets up the background model under the RGB color space. Firstly, the Codebook background model based on the characteristics of luminance and chrominance separation of YUV color space was built in this thesis. Secondly, The average of the Y component in the codeword was used to redefine the brightness range aimed at the problem of the irrational brightness range definition of the background model and made the brightness range update with the changing background in this thesis. Finally, a background updating method which combined a two-layer Codebook model with a short-sliding window was proposed to update the background in the process of moving object detection in order to improve the anti-jamming capability in this thesis. The improved method could update background effectively and improve the precision at the same time.2. Various tracking methods was implemented and compared in this thesis, and particle filter algorithm was introuduced into the tracking module due to it is applicable to any state space. The particles' weight calculation method which combined the location prediction and template match was put forward and it improved the tracking accuracy and efficiency under the premise of insuring the particles' diversity.3. The judgment method of moving objects' abnormal behaviors was analyzed. And the detection method of run speed increase, fall and hover were achieved through extracting the characteristics of the moving target(trajectory, mass center, the ratio of high to width of rectangular box which belongs to the moving target) and designing reasonable judging methods.4. An abnormal behavior detection system which could track the moving target and analyzed the abnormal behavior accurately was implemented. And it could find the abnormal behavior of video moving object in the scene and send alerts timely.Finally, the research content was summarized and the problems which existed in the research was put forward, and the idea of solution which laid a foundation for subsequent research work was raised.