Research On Object Tracking Techniques Based On Particle Filtering

Object tracking has always been the important direction in the image processing field, because this research is close to our real life, and therefore, has always been the concern of researchers. On object tracking lots of efficient algorithms ha ve been proposed, which are mainly concentrated on the Kalman filter and Mean-sift approches. However, Kalman filtering based methods are limited by the system of linear Gaussian system. At meanwhile, the methods of Mean-sift have certain limitations for fast moving and on non-Gaussian environment. Especially, when occlusion occurs on the tracking process, the system efficiency would be decreased. Condidering on the problem of these algorithms, the particle filtering has drawn much more attentions, since this method is not limited to the system model and can effectively deal with the occlusion problem.However, there are also a series of problem need to be resolved, such as particle filter needed lot of particles to track the object, in the process of tracking the samples of dilution and so on. Therefore, in this paper we research particle filter tracking problem as the center to expand work. Research and innovation work is as follows:(1)An adaptive multi-feature tracking algorithm based on particle filter is proposed. In the analysis of a single partic le filter performance characteristics based on the proposed tracking algorithm, we proposed a new method using color features and gradient features are integrated into the framework of particle tracking, improve the effectiveness of the object tracking algorithm.(2)In this paper, a novel combined particle swarm optimization with particle filter algorithm is proposed for the object tracking problem. In the process of particles filter sample impoverishment is an inevitable issue, through particle swarm optimization was embedded into classical particle filter framework to achieve better search results through constantly iteration. However, PSO-PF too easy to fall into local optima, according to compare if the global optimum values updates changes some particles location to jump the possible local optimum. Experiment results show that the proposed method maintains the diversity of particles and uses fewer particles improve the tracking efficiency.(3)An adaptive tracking algorithm uses particle distribution model under Rayleigh distribution is proposed. The algorithm of particle filter has a most prominent problem is not efficient, because in the process of tracking it requires a large number of par ticles to estimate of the real state. In this paper we follow Rayleigh distribution function, and the parameter of Rayleigh adaptively to the model to realize the particles more beneficial for the target tracking. And in the process we dynamic judge the tracking result to adjust the number of particles. In this way the efficiency of tracking has improved and the effect also has enhanced.Finally, the summary of this thesis is made, and further research is discussed as well of this study, so future research direction is pointed out.