Moving Object Tracking Algorithm Research Based On Color And Particle Filter

Video objects tracking is a core problem in the fields of computer vision, which has a wide range of applications in malitory guidance,visual surveillance,visual navigation of robots,medical diagnose weather analysis and so on. The research of video image processing technology focus on how to detect and track moving targets quickly and accurately.Based on the summary and analysis of the existing method of moving target detection and tracking ,the thesis focus on the moving target detection and tracking technology in a static scene.For the tracking problems in complex environment,The thesis focus on the solution to the problems such as observation model of target tracking, inter-block between targets and the resampling of particle filter algorithm , and it proposed corresponding solutions to such issues. The main contents of this dissertation are summarized as follows:Firstly,analyzed the the existing tracking algorithms,with an emphasis on elaborating the principle of Bayesian filtering theory, including particle filter and Kalman filter algorithm,discussed the framework,advantages and disadvantages of the algorithm.Secondly ,for the problem of moving targets in natural environment are vulnerable to the background interference, an improved color distribution model was proposed, and brought it to the oparticle filter observation model.Thirdly,faced with the issue of poor particle in traditional particle filter algorithms, we introduced a resampling particle filter algorithm which is based on diversity sampling technology.The algorithm added a link of diversity sampling after traditional re-sampling algorithm. To find related particles according to a uniform distribution in the neighborhood area of particles after resampling, so that the particles will not converge to one point and increase the particles diversity, In addtion,the algorithm achieved the purpose of solution to particles depleted problems and solved the particle tracking failures due to depleted particles as time increased.Fourthly,the thesis analyzed the occlusion problem for targets tracking and brought up an improved occlusion detection processing algorithm, which determined the target state through Bhattacharyya results ,if the target was hidden, it updated current frame state by using the template of the previous frame ; when the hidden was over, then recovered the target template updates.Finally, this method was applied to the video image test in the natural environment. The experimental results showed that this approach can track the target accurately and achieved a higher robustness when the target was severely hidden.