Research On Moving Object Detection And Tracking Algorithms For Intelligent Visual Surveillance

As a frontier research involving many subjects, such as computer vision, image processing, intelligent control, the intelligent visual surveillance technology has become one of the hot issues in the field of computer visual research at home and abroad, and the research for it has a very important scientific meaning and extreme broad application prospects. The technology of moving object detection and tracking, as one of the key technologies of intelligent visual surveillance, the challenge and the technical complexity of its problems which are to be solved attract and encourage many computer vision researchers into the research ranks, and the performance of the annual innovative algorithms proposed is significantly improved. However, in practical applications, moving object detection and tracking are still having many challenges and difficulties, which are worth our further analysis, research and discussion. Based on these studies, this paper carried out the research for the key technologies of intelligent visual surveillance.First, the paper briefly reviewed the research status and development trend of the intelligent visual surveillance technology, and deeply discussed the key technologies and difficult issues of it and clarified the topic’s background and significance. Next, the paper studied the basic theoretical knowledge of digital image processing in detail, including image gray processing, color model, image de-noising, histogram equalization, mathematical morphology theory, and so on. The study of these basic theories, made a foundation for the in-depth discussion of the moving object detection and tracking technology.Secondly, the paper analyzed the three classic moving object detection algorithms:the optical flow method, the background subtraction, and the fame difference method. After the comprehensive comparison of the advantages and disadvantages of these three methods, we proposed a symmetric differential object detection algorithm combining the background subtraction and the spatial temporal entropy. This paper selected the standard video test sequences named "hall_monitor" and "Intelligentroom" to verify the effectiveness of the algorithm. Compared with the traditional symmetric difference, the improved symmetric difference algorithm has a better de-noising ability, and the integrity of the object detection is greatly improved. What’s more, it can avoid effectively the "ghosting" phenomenon in the object detection, and greatly enhance the efficiency of the moving object detection, and also can overcome the influence of environmental noise such as the illumination changes.Then, from the perspective of the principle of the algorithm, the paper introduced four types of moving object tracking algorithm, and discussed the current object tracking technical difficulties. On the basis of these theories, the paper in-depth studied two classical algorithms:the mean shift algorithm and the particle filter algorithm. The mean shift algorithm has a better real-time, but the robustness is poorer; the particle filter algorithm has a stronger tracking accuracy, but the large amount of calculation and the degradation of particles constrain its real-time. Considering their advantages and disadvantages, this paper proposed a particle filter algorithm for object tracking fusing the mean shift algorithm. Using the clustering analysis of the mean shift algorithm, the new algorithm can improve the efficiency of particle sampling, and inhibit particle degradation effectively; what’s more, adding the object template update link in the implementation process, the algorithm can perfectly solve the object occlusion problem, and greatly reduce the amount of calculation of particle filter under the premise of guaranteeing the tracking performance, so that it is applied to real-time surveillance system becoming possible.Finally, a simulation experiment was conducted using the video test sequence "CAMERA1" of PETS2001in MATLAB7.10environment. Based on analysis of the results, the new proposed algorithm can achieve equal tracking performance with conditions of less particles and execution time lower than one third of the traditional particle filter. Having higher robustness, accuracy and real-time, it not only can resist the interference of illumination changes in terms of background factors, but also can perfectly handle complete blocking issues under complex backgrounds.