基于遗传粒子滤波器的运动目标实时跟踪

Real-time Moving Object Tracking Based on Genetic Particle Filter

提出一种基于遗传粒子滤波器的运动目标跟踪算法,它将Boosting算法和遗传算法引入粒子滤波器,构建了遗传粒子滤波器.该方法首先利用背景信息和目标信息建立特征分类器,将分类器的输出结果作为粒子滤波系统观测的重要信息,进行粒子权值的计算;并在跟踪过程中不断更新特征分类器,从而自适应地更新粒子的权值.为了提高算法的实时性,将遗传算法引入到粒子滤波器,在保证粒子滤波器精度的前提下,减少粒子数目,从而降低算法的运算时间.实验结果表明,所提算法可以根据背景信息的不同自适应地选择特征,在遮挡、形变及背景干扰等情况下,依然可以很好地对目标进行稳定的实时跟踪.

This paper presents a moving object tracking algorithm based on genetic particle filter which is constructed with boosting algorithm and genetic algorithm. The object information and background information are used to construct feature classifiers, the output results of these classifiers are taken as the important observation information for the particle filter system and are used to calculate particle coefficients. These classifiers are updated during tracking so that the particle coefficients are updated adaptively. Genetic algorithm is introduced into the particle filters to improve the real-time ability of the algorithm. On the premise of guaranteeing the accuracy of the particle filters, the number of particles is considerably reduced and the processing time is decreased. The experiment result shows that the proposed algorithm can adaptively select features according to different background information, and can carry out stable and real-time tracking event if covering, deformation and background interferences exist in the environment.