Improvement of Moving Objects Tracking via Modified Particle Distribution in Particle Filter Algorithm

Object tracking is an important issue in machine vision, which has many applications. A tracking method is particle filtering that is based on Monte Carlo techniques. This method is based on random sampling of a probability density function and estimating the desired variable using samples weight. In this paper, particle filter algorithm is implemented by considering the color histogram model as the existing observations. In order to investigate the particle filter performance, a comparison between this technique and the mean shift method is presented which reveals that the proposed method has better performance. A problem associated with particle filter method is degeneracy phenomenon. By modifying the particles distribution, we avoid increasing in the particles weight variance, which is the main reason of degeneracy phenomenon. Applying the proposed method on the standard databases demonstrated better results. Further, since in the proposed scheme the particles are distributed in improbable areas, if any occlusion occurs, the probability of the target missing decreases and the target tracking will be done more successfully.