Target Tracking In Video Sequences Based On Particle Filter

Video tracking is an important research field of computer vision for its wide application requirements and prospects in many industries. The main task of video tracking is to track a moving target in video sequences so that we can get the position, velocity, trajectory and other parameters of the target. Recently, the method of particle filter based on Bayesian recursive estimation framework provides new solutions to video tracking. However, it is a considerably challenging research to make a stable and accurate video tracking by particle filter in actual target tracking, for the reason that the illumination, target profile, complex background, occlusion and other factors increase the difficulty of tracking. In this paper, corresponding solutions will be found in the following aspects.Target feature extraction is not easy to control in actual complex environment of target tracking and the tolerance of various characteristics’likelihood observation to noise and other interference. Combined with kernel particle filter that has parameter-free estimate ability, a new particle filter tracking algorithm is figured out in this paper. When tracking a target, our algorithm can automatically and dynamically adjust the weight of different features, combine the information of feature observation adaptively, improve the ability to identify the target in complex view field, provide the parameter-free kernel based on the observed sample then estimate the weight of particle, enhance the use of observed sample’s information and improve target tracking performance. In addition, it finds a new tendency predicting method with sliding weighted window which can be used to predict target’s moving trend and better the spatial distribution of particles so that enhance the target tracking capabilities of the particles.Normal particle filters use suboptimal importance sampling function, this can easily result in particle degradation and Impoverishment, and cause the decline of tracking precision and shrink of tracking space, and reduce the performance of target tracking. So it’s applicable to introduce particle swarm optimization into normal particle filter and improve it by niche technique, to make particles form a multi-population distribution when they move into high likehood area of target. This can boost particles searching ability to target state space, thus improve adaptivity to the change of dynamic target state. This paper integrate two kinds of particles effectively, that is, particles of niche optimization and particles of particle filter, to introduce a new algorithm for particle weight correction, which finally realize a new Niche-PSO particle filter for video target tracking.Face profiles is various in different illumination and different postures. Occlusion, background changing, complicated moving of face; all above bring huge challenges to face tracking. Integrated with two good performance particle filter algorithms mentioned before, this paper introduced a new face tracking particle filter algorithm. It integrated adaptive observation of multi-feature objects and transfer ability of multi-population, high likehood particles. It improves the distribution of particles in space, and the computational method of particles’weight. It thus enhanced particles’ tracking ability to target state, and enhances the algorithm’s precision. In addition, by integration of a new target template updating strategy proposed by this paper, this method can deal with the situation when face was occluded in large area, thus further improved the robustness of our face tracking algorithm.