Target Tracking Based on Particle Filter and Mean Shift

Abstract

Robustness and real time are just two pursuing goals in visual tracking area. The main advantage of particle filters is its robust performance, but it can't be applied in critical real time requirement fields due to its long time consuming. Although mean shift algorithm have the advantage of real time, tracking drifts will occur if the video contain complex scenarios, that is to say, the disadvantage of mean shift algorithm is its poor robustness. Infrared target tracking algorithm used in imaging guidance system must have the advantage of both robustness and real time. Aimed at the fact, a novel tracking method tightly coupling particle filter and mean shift algorithm is proposed in this paper. Compared some existing tracking algorithm using both particle filters and mean shift, the tight coupling in the paper means that a five parameter set (2-D central location, width, height, and orientation) is utilized to fully describe a possible state of target, and the parameter set not only works as a particle in the particle filter algorithm, but also can be revised by the mean shift algorithm, other than the 2-D central location only. Experimental results show that the proposed tracking method is robust and in real time, even if videos contain complex scenarios.