Research On Multi-target Tracking Algorithem Based-on Multi-modal Particle Filter

As an important research areaof computer vision, object tracking has attached great attention in the field of military and national defense. In modernized wars, trackingand making strikes of air-craftsaccuratelybecomes the key to defeat the enemy. Because of the loss of diversity among particles in re-sampling process, the performance of traditional particle filter tracking algorithm is not ideal withdramatic changes oftarget attitudes and sudden disappearance of the target, therefore it is very important maintain the diversity of particles as well as reduce the amount of calculation.This paper proposes a multi-layered particle filter tracking algorithm to deal with in dynamic background, particles are divided into the principal particles and subordinate particles,the principle particles are optimized by subordinate particles to improve the diversity. A multi-modal particle filter multi-target tracking algorithm is proposed, for the reason that the increasing number of targets would exponentially increase calculation, and Mean Shift algorithm is used to significantly reduce the amount of calculation. In this paper, the proposed algorithm has been proved feasible through experiments. This article includes the following works:(1)In this paper, the traditional particle filter is set as a starting point, theoretical analysis and programmed to achieve the traditional particle filter target tracking algorithm, based on the results, the losing of particles diversity in re-sampling is found.(2)Improving the diversity of particlesby increasing the number of particles,according to the comparative of experimental data, only increase the number of particles will exponentially increase calculation and improvelittle diversity of particles.(3)Presents a multi-layered particle filter tracking algorithm in dynamic background, particles are divided into the principal particles and subordinate particles,subordinate particles are used to maintain the state ofprincipal particles and optimizeprincipal particles as the same time. In re-sampling, only principal particles are updated while subordinate particles remain unchanged, at last, principal particles areused to estimate the state of targets. This improvement improves the diversity of particles to some extent,has a good effect in response to dramatic changes in target attitudes and the temporary disappearance of the target. Experimental results show that the proposed algorithm is better than the basic particle filter algorithm(4)Proposes a multi-modal particle filter multi-target tracking algorithm fitting in Mean Shift and shows the detailed process of establishing methods and algorithms of multi-objective joint model, and based on this method, experiments for single-target,multi-objective, static background, dynamic backgrounds, and other statesare made, the experimental results show that the multi-modal particle filter multi-target tracking algorithm proposed in this paper can track the targetaccurately, has a good effect in rapid changesand partial cover of the target.