基于目标模板库的多任务目标持久跟踪

Long-time multi-target tracking based on the target template library

针对长时间跟踪造成的信息丢失问题,提出了一种借鉴人类视觉记忆机制构建目标模板库的算法,该方法能在跟踪中记忆有用目标信息,实现持久稳定的跟踪.首先采用多任务跟踪法把视频序列分成多个子任务进行多线程分块局部跟踪,然后采用模板匹配和特征融合下的粒子滤波先后进行粗略跟踪和精细跟踪;最后把跟踪结果纳入目标模板库中更新跟踪系统.实验表明,此算法具有较好的鲁棒性和稳定性.

Aiming at the information-lost problem caused by long-time target tracking,this paper proposes a novel algorithm by imitating human visual memory mechanism to construct a target template library which saves useful information and realizes a stable and robust tracking.Firstly,a multi-task tracking method is adopted to divide the frame into several tracking processes which are composed of local subtasks.Second,using the template match and particle filter of feature fusing to accomplish rough tracking and precise tracking.Finally,the preliminary tracking results are put into the target template library to update the whole tracking system.Experimental results demonstrate that this algorithm has a good robustness and stability.