基于遗传重采样的粒子滤波视觉跟踪算法

Visual Tracking Algorithm Based on Particle Filter of Genetic Re-sampling

针对传统粒子滤波算法的粒子退化和枯竭问题,提出了一种基于遗传重采样的粒子滤波视觉跟踪算法。该方法将遗传算法的进化思想引入到重采样过程中,根据有效粒子数选择每代粒子中较优个体,对在预先设定的阈值下的粒子进行交叉运算,利用MCMC方法实施变异,从而解决了粒子的退化和枯竭问题。另外,采用自适应模板更新技术以解决跟踪中目标表观变化的问题。实验结果表明,该算法是可行的。

Aimed to crucial issues in traditional particle filter（PF） of removing the degeneracy phenomenon and alleviating the sample impoverishment problem,an improved particle filter tracking algorithm with genetic evolution theory is proposed.In the addressed algorithm,the issues is solved by applying the selection operator of genetic algorithm to choosing the optimal samples based on effective particles,by the crossover operation implemented to the samples under the predefined threshold,and by the MCMC mutation operation to all new particles.Furthermore,considering the target appearance changes,the template is updated adaptively for keeping the accuracy in the scenario of tracking.Experiments show the feasibility of the proposed method.