摘要
针对粒子滤波算法中粒子退化导致跟踪效果差的问题,将均值漂移(mean shift,MS)算法融入到粒子滤波(particle filter,PF)算法中,提出了基于MS重要性采样的粒子滤波目标跟踪算法。该算法根据MS算法的核函数原理,利用目标颜色分布建立目标参考模型;基于粒子滤波的计算框架,在MS算法的迭代寻优过程中,对粒子进行确定性搜索,使其收敛到目标候选模型的局部最优点,完成对目标模型的重要性采样,有效解决了粒子退化问题,同时减少了粒子数目,增强了目标跟踪的实时性。跟踪实验结果表明,该算法的跟踪性能优于传统的粒子滤波算法。
To solve the problem that particle degradation results in bad effects of tracking in particle filter algorithm, Mean Shift （MS） algorithm is integrated into particle filter （PF） algorithm and a particle filter algorithm based on MS importance sampling for target tracking is proposed. Firstly target reference model is constructed based on MS kernel principle and color distribution of target in this algorithm. Then deterministic search for particles is done in the iterative optimization process of MS algorithm based on the computational framework of particle filter, converging particles to local optimized points in target candidate model. At this time, the importance sampling on target model is completed and the problem of particle degradation is solved effectively. Furthermore, the number of particles is reduced, which is helpful to enhance real-time performance. The final tracking experi- ments show that the proposed algorithm is superior to traditional particle filter algorithm in terms of tracking performance.
出处
《计算机工程与设计》
CSCD
北大核心
2014年第3期909-913,共5页
Computer Engineering and Design
