Research On Moving Target Detection Algorithm For Video Surveillance

As public safety is facing with more and more challenges, the traditional videosurveillance system can not meet the requirements, intelligent video surveillancesystem arises at the historic moment. Intelligent video surveillance system is anintegration of multiple techniques such as computer vision, image processing,intelligent control and etc; it allows to perform effective real-time monitoring. Asthe foundation of intelligent surveillance system, Moving object detection is a coretechnique, it is the basis for the subsequent target tracking, recognition and othertechniques, so it has very high research value and promising future. The study isbased on moving target detection under a stationary camera.In this paper, the embedded system is adopted to control video acquisition andtransmission. The system is built upon ARM9processor, and the video captureterminal adopt CMOS OV9650camera. Video information is acquired throughVideoLinux2interface, and then transmitted to PC through the RTP/RTCP protocol.Moving target detection is mainly performed in PC under Visual C++6.0compilingenvironment, while taking advantage of the open source vision library OpenCV1.0.Firstly, three traditional algorithms for moving target detection, namely, optical flowmethod, inter-frame difference method and background subtraction method, arepresented and discussed. Then, their principle, advantages, disadvantages andapplicable scope are analyzed in detail and corresponding simulation is performedand analyzed. Finally, a number of improved algorithms that’s based on the threeclassical algorithms are discussed, among which the background modeling-baseddetecting algorithm is mainly discussed.In order to improve the poor performance of the original codebook model inreal-time background modeling and efficiency of algorithm, this paper presents animproved algorithm. The algorithm effectively combines background modeling andmotion detection. This model is mainly realized by removing redundant elements ofcodebook, it uses a variable to replace the first and last access time of original codemodel. This method ignores the brightness information in order to save memory.Compared with the original codebook model, the proposed method effectivelyimproves the processing speed and takes up less memory. In addition to updating thebackground to achieve real-time detection of moving targets through the integration of Gaussian mixture model. Compared with other moving target detection algorithms,his algorithm can update the background model in real time, less noise and nointernal voids in video frame. In addition, it plays a very good inhibition effect onthe environmental change and shadow interference, the results show the algorithmha-s better robust, real-time and accuracy.By integrating video image acquisition module, transmission module, imageprocessing module, moving target detection module, the detecting system for movingtargets is realized. After system testing, it can achieve real-time moving targetdetection and achieve the expected goal.