Research On Multi-moving Objects Tracking Technology In Video Based On Particle Filter

Moving targets tracking technology in video is the research focus of computer vision field ,andis the key technology of moving object identification, unusual events detection, robotauto-navigation and etc. Because in video the cases of targets defilade, cross over and separationgenerally exist. The issues of multi-target tracking in video have always been the research focus anddifficulties. The framework of detecting before tracking is still used in this paper, discussing theissues of moving target detection and tracking separately. The main contents of this paper areintroduced as follows:Merging Background Subtraction method and Frame Subtraction method, the incrementationsof neighbor image frames are used to update the background model, and adjust the rate ofbackground update in the detection process. So that the adaption of background changes ofdetection algorithm, the accuracy and integrality of background detection are improved. Theincrementations can be used in other target detection algorithms directly or flexibly.Kalman Filter and the Particle Filter are commonly used in the video objects tracking. thispaper analyzes the application of the video target tracking which used the Kalman Filter and theParticle Filter. By the introduction and analysis of the movement target model, we discussed theprinciple of Kalman Filter and carried on the simulation analysis. The simulations of ExtendKalman Filter, Unscented Kalman Filter and normal Particle Filter are executed.The Computation of the Particle Filter is the key obstacle to the application, the conventionalRao-Blackwell and the Rao-Blackwell based on particle filters reduce the number of particles andreduce the iteration number of the algorithm by using the marginal distribution of particle filter andthe random sample of Kalman correction technology. This paper achieves the video objects trackingby the using of Rao-Blackwellization Particle Filter based on the data association framework MHT.This algorithm not only achieves the video objects tracking successfully, but also realizes theeffective judgment of the targets’birth and death.