Research On Object Tracking Algorithm Based On Particle Filter In Complex Background

Video target tracking is a complex subject which integrates pattern recognition,image processing, filtering theory, probability theory, stochastic processes and otherdisciplines. As the key technology of the computer vision, it has been widely used inmilitary, industrial, civilian and other fields. Although after the researches of manyscholars, people have achieved some fruits in video target tracking, but it can notmeet the requirements of the higher level.Aiming at the problems that the target can not be tracked accurately undercomplex background, this paper optimized the tracking algorithm by using themethod of combining intelligent algorithm with particle filter and establishingobservation model of multi-feature fusion, thus overcame the problems of targets areblocked partially under complex background. There are two aspects in this paper:Firstly, Aiming at the problems of particle degradation and particleimpoverishment in particle filter, this paper proposed a particle swarm optimizationparticle filter based on chaotic algorithm. Drive particles move to the high likelihoodarea through particle swarm optimization, and make particles escape from localoptimum area timely through chaotic disturbance and searching, make the particlesfind the global optimum position, so that the particles move closely to the globalbest position, thereby increasing the useful particles and the diversity of particles, itcan inhibit the particle degradation and impoverishment effectively. The simulationresults show that the new algorithm this paper proposed could remarkably improvethe estimation accuracy compared with the conventional particle filter and thetraditional swarm optimization algorithm.Secondly, Aiming at the problems that it is very difficult to track the target whenextracting only one feature, and that the tracking algorithm will fail when the targetis blocked, this paper proposed a new particle filter algorithm which combinesmultiple features. It overcame the problems that the target will loss when it isblocked partially through establishing observation model that integrates featuresboth color and texture, determining the dynamic system noise variance adaptively,increasing the particle activity range to overcome the problems that the target willloss when it is blocked partially. The experimental results show that the method of this paper proposed is very robust, and it can deal with the problems that the targetwas blocked partially under complex background better to achieve trackingaccurately.