Application Of Mean Shift Algorithm In Video Target Tracking

Video moving object detection and tracking technique is one of the key research topics in Computer vision level. It is the most efficient, accurate and rapid visual rackingtechnology direction. Video moving object detection tracking technique is one core technology in Courseware recording system. Its main function is to track the location and tracking of students and teachers image. But Image tracking and positioning quality will have a significant impact on image quality.The mean shift algorithm is a kind of non parameter estimation method. It is carried out mainly through the kernel function. The main advantage is that it need no priori knowledge, and it support for real-time tracking and fast convergence. In recent years, it has already got the attention and research interest of many scholars. In this research, the mean shift technique is seen as the core content for the study of the correlation tracking algorithm. Then further optimize the tracking performance. The main research contents of this paper and the innovation mainly includes,1. For Mean shift algorithm, this paper focuses on studying mean shift algorithm’s characteristic which is designed by the probability distribution and kernel histogram algorithm. Then, pitches in the bad background color of robust caused by the probability distribution in the traditional algorithm. Finally give a optimized tracking algorithm by the target detection target. The experimental information displays the target can effectively be tracked through this algorithm when the scene contains the great elements which are similar to the target color.2. The particle filter algorithm are analyzed and researched in the basic Bayesian filter theory. Firstly, this paper comprehensively states the particle filter algorithm operation mechanism and the application in image target tracking. Then explains the outstanding problems of robust tracking performance which appeared in the process of the traditional particle filter object description through the single feature. Finally, it provides the target model for the particle filter algorithm characterized by the target color and motion.3. There are some problems in the traditional Mean shift algorithm and Particle filter algorithms, this paper mentioned an algorithm that is Fusion of mean shift particle filter algorithm. First of all, improve the Mean shift algorithm through the object description model. Develop from the traditional algorithm which uses H components distribution map to the new algorithm which uses H, S component integrated color probability distribution map to describe target. Then put the target movement information into the model escription to establish the target model through the Movement probability distribution map. After that, put the improved algorithm into the Particle filter framework. Use it to iterate all particles before the Particle weight updating, redistribute of sample particles to move to the target of maximum posterior probability density function direction. In this way target state can effectively be estimated. At the end of this paper, the full text of combing and summarized, summed up the fusion of Mean shift particle filter algorithm to solve the problems existing in the original recording and broadcasting systems, and tracking analysis results, to achieve the desired objectives through three groups of experiments.