Abstract

To improve the accuracy and the real-time of the multi-target tracking when the background changes and occlusion, we propose a multi-target tracking algorithm based on fusion of Mean-Shift and Particle Filter. Firstly, obtain the background image model by inter-frame difference method and detect the moving object through the model, update the back-ground real-time by impulse response during the tracking process. Secondly, based on the particle filter, the mean shift was introduced to redistribute random sample particles, in which particles move toward the maximal posterior kernel density estimation of target state, the weights of particle samples are updated as the mean shift iterative operating. Finally, obtain the position of the object in next frame, which used mean shift iterative operation and the relationship of moving object in before and current frame. Experimental results show that the algorithm we proposed obtains more accurate tracking results and better robustness when background change and occlusion.

Paper Details