Research On Object Tracking Algorithm Based On Particle Filter

The video frequency movement target tracking is the most active research subject in computer vision domain at present, as an interdisciplinary frontier technology, it fused on many kinds of different domain theoretical knowledges such as the imagery processing, the pattern recognition, and the artificial intelligence and so on; it widely applies in the military and the civil aspect. Therefore, the domestic and foreign scholars have launched the massive research work in the video frequency movement target tracking domain, and obtained some results. But it still had some difficult problems in target tracking, such as deformation, mask and fast-moving about object as well as strong misalignment, non-Gauss’s state estimation question. The particle filtering has excellent robustness in target tracking with various forms as well as in the nonlinear motion and measurement model; it is an effective way to solve such problems, but it has problems of particle degradation and scarcity which is cause by the resampling. In view of the above question, this paper mainly focuses on the particle filter technology and intelligent optimization algorithms for video target tracking study in order to improve the performance in real-time, robustness and accuracy of target tracking method. The work completed in this article is as follows:1. Proposed a CIGA-PF algorithm to solve the degradation problem when non-Gaussian posterior probability distribution was resampled. The algorithm can prevent the premature phenomenon of the genetic algorithm to improve the efficiency of problem solving and avoid the algorithm into a local optimal solution. The algorithm can also effectively solve the problems of particles degradation and deprivation in the particle filter to ensure the diversity of particle.2. Proposed an IPSO-PF algorithm when improved the resampling step. The algorithm introduced isolation niche technology into the resampling process, so that it can improved the particle resampling process through the evolution of the various sub-groups by flexible control. Compared experiments with the SPF algorithm showed that the sampling location of the algorithm is relatively fixed, the sample distribution is uniform and it can be achieved the requirement the particle diversity.3. Finally, two improved particle filtering algorithm were used for video object tracking and compared with the IGA-PF and the SPF algorithms. Experimental results show that CIGA-PF algorithm had stronger robustness and be better to deal with the impact of the target occluded in image sequence, it had more effectively when tracking the target; on the other hand, IPSO-PF algorithm was better in real-time and applicable to the target in the motor state.