联合Mean-shift与粒子滤波器的实时视频目标跟踪算法及实现

摘要

粒子滤波器由于摆脱了高斯分布的约束条件,已经成为一种主流的、面向目标的非线性运动跟踪算法,广泛应用于视频压缩与检索、智能视频监控、智能人机交互等领域,其缺点是计算复杂度高、计算量庞大,无法满足实时应用的需求.针对粒子滤波器在计算量、实时性及粒子退化方面存在的问题,提出了将Mean-shift算法嵌入粒子滤波器,对重要性采样分布进行优化,以较少的采样粒子实现视频目标跟踪.仿真实验结果显示,联合Mean-shift的粒子滤波算法在目标跟踪过程中具有较好的实时性与鲁棒性.%Due to get rid of the constraint condition of Gauss distribution, particle filter has become a mainstream, target-oriented nonlinear motion tracking algorithm, widely used in video compression and retrieval, intelligent video surveillance, intelligent human-computer interaction and other areas, the drawback is the high computational complexity and huge amount of computation, can not meet the needs of real-time applications.In this paper, the problem of particle filter in computational complexity, real-time and particle degradation is proposed.The Mean-shift algorithm is embedded in the particle filter, and the importance sampling distribution is optimized.Video target tracking is achieved with less sampling particles.The simulation results show that the joint Mean-shift particle filter algorithm has good real-time and robustness in the process of target tracking.