一种结合Mean—shift和粒子滤波的视频跟踪算法

提出一种结合均值偏移算法和粒子滤波理论的视频跟踪算法，解决了目标旋转、部分遮挡和运动速度过快的问题．通过均值偏移对粒子滤波中的粒子集进行进一步寻优，增加了粒子的有效性，极大减少了粒子采样的数量，且解决了经过多次重采样后粒子多样性降低的问题．新的粒子通过与观测值之间的巴氏系数来决定重要性权重．实验证明：本算法可以完成实时地对视频目标进行部分遮挡以及目标旋转下的跟踪，具有较强的鲁棒性．

A Video Tracking Algorithm with Fusion of Mean-shift and Particle Filtering

An algorithm for effective video tracking based on fusion of Mean-shift and particle filtering （PF） is proposed. The research efforts are made to tackle the problems with the target rotation, partial blocking and fast-moving, etc. Using Mean-shift in particle filtering is designed to further optimize the particle set, thus increasing the effectiveness of particle identification, reducing the number of particle sampling, and mitigating the drawbacks found in particle diversity reduction arising from multi-sampling process. Bhattacharyya factor is applied to determine the priority weighting between the new particle and measurements. The algorithm is put to real-time target tracking test on partial blocking and target rotation, manifesting the adequate efficiency and robustness.