使用特征融合的多级别MS-PF目标跟踪算法

Multi-level MS-PF Tracking Algorithm Using Feature Fusion

提出一种改进的结合均值漂移与粒子滤波(MS-PF)的跟踪算法,通过分级策略将目标跟踪分为直接跟踪、微调和搜索三个级别,实现了均值漂移和粒子滤波的动态结合.针对传统跟踪算法特征单一的缺陷,在目标跟踪过程中自适应的融合了颜色和纹理特征,同时引入粒子群优化算法对粒子滤波进行优化.实验结果表明,采用分级的MS-PF算法能对粒子的产生和数量进行严格控制,提高了算法的实时性和通用性,在复杂环境中,尤其是在光照发生变化时,基于特征融合的思想使得算法更具鲁棒性.

An improved tracking method which combined with the mean shift and particle filter （ MS-PF ） is proposed. It implements a dynamic combination of mean shift and particle filter by using grading strategy, tracking will be divided into three levels include di- rect tracking, fine adjustment and searching according to specific cases. The tracking process is based on Adaptive fusion of color and texture features, and introduces particle swarm optimization algorithm to optimize the particle filter. The experimental results show that using the graded MS-PF algorithm can implement strict control on the generation and number of particles, wh/ch improves real- time capability and universality of this algorithm. Meanwhile, feature fusion strategy makes the algorithm more robust, especially in some complex environments.