The Research On Moving Target Detection And Tracking Method In Intelligent Transportation Systems

Intelligent Transportation System (ITS) is the project which develops rapidlyin our social construction, but ITS is one of the region which still need to improveurgently, meanwhile, ITS is the hot area of research in the different countries. Thealgorithm of moving target detection and tracking based on video is the important partof the ITS and the basic technologies. However, the algorithm of moving targetdetection and tracking faces the enormous challenges of the complex environment, suchas occlusion and variation of target, in practical application with video.Aimed at the problem above, this paper proposed a series of developed methods,which tackle well with the problem of target detection and tracking after theinvestigation of the methods of moving target detection and tracking at present. Thetracking algorithm proposed is mainly designed to deal with the problem of occlusionand variation of target. In order to overcome the problem of complete occlusion oftarget, a target search strategy is proposed, which is succeed in the application of thetarget tracking. There are the main works finished as follow:Firstly, the moving target detection. In order to solve the problem that fault oftarget with traditional Gaussian Mixture Model, the algorithm of moving targetdetection that combined the inter-frame difference method and Gaussian Mixture Modelis proposed. Therefore, the foreground objects are extracted from background moreapparently, and the problem that fault of target is solved.Secondly, the algorithm of target tracking. The blocked tracking algorithm ofMean-shift is proposed. The target is divided into some similar sized blocks, and theposition of the small blocks are predicted by Kalman filter. The tracking algorithm ofMean-shift is respectively used in tracking the small blocks at the position predicted byKalman filter. The effectiveness of target blocks are estimated by tracking detector andthe invalid blocks are screened from the target. Thus, the problem that the reduction of tracking performance caused by target incision is solved. Much more spatialinformation can be detected by the method, so the limitation of Mean-shift algorithm ismade up and the robustness of tracking method is increased.Thirdly, under the condition of occlusion. The problem of occlusion can be separateinto complete occlusion and partial occlusion. Both the problem that variation of targetand partial occlusion, belong to the problem that partial information loss. Therefore, theblocked tracking algorithm of Mean-shift is straightly used to track the target under thecondition of partial occlusion. When the target is occluded completely, the normalizedcross correlation image matching algorithm based on adaptive step size and the methodof foreground detection are used to search the target information out of occlusion, andthe tracking is successful under the condition of occlusion.According to the methods above, these algorithms is programmed by computertested with the videos. The experiment results indicate that the proposed algorithm canefficiently and accurately track target under the condition of occlusion and variation oftarget.