Improved Image Sequence Moving Target Detection And Tracking Algorithm

Visual information is one of the most common and basic content in the media information.Therefore,how to process information efficiently and serve the people well had been a serious challenge.The rising of computer vision makes up for the shortcomings of the human effectively,it can use computer to simulate the biological vision for processing data.In the computer vision field,target detection and tracking are the popular research topic.It has been widely used in a plurality of military defense,intelligent surveillance,human-computer interaction, intelligent transportation and other fields.There are many scholars committing to study of target detection and tracking. In this paper,based on the existing research results,it pointed out their deficiencies and gave some advices on how to improve the basic algorithm correspondingly. This article ' s main work was as follow:In the target detection, it mentioned three conventional detection algorithms and summarized their advantages and disadvantages. By considering the fuzzy phenomenon in the background initialization of traditional background subtraction, a novel background initialization method on combining mean modeling and median filtering is proposed. For the reason that updating background timely and quickly, an adaptive updating coefficient was proposed. The method combing five-frame diffidence and two-dimensional maximum entropy was put forward in order to reduce the serious phenomenon of "empty" and "shadow" in the three- frame difference.Above all,it also combined two improved algorithms with logic operation. No matter in the slow motion or in the fast motion, the experiment result showed that the new algorithm had better detect result.In the target tracking fields, Meanshift and particle filter are powerful and reliable tools. Meanshift is an efficient and fast algorithm. However, it is more sensitive to the illumination changing and occlusion. The particle filter algorithm has good robustness to occlusion, but the tracking speed is slow caused by large of computation. Owing to the weakness of using independently,this paper presented the algorithm of integration based on Meanshift and particle filter. With the help of Meanshift algorithm. it determined the sampling particle number, the state estimated by the particle filter was offered to Meanshift. In the experimental results, the improved algorithm was obviously better than the traditional algorithm.