提出了基于序贯蒙特卡罗概率假设密度滤波器( SMC -PHDF)的视觉多目标跟踪算法.W4算法对观测场景进行背景建模和运动目标检测,获取可能目标在观测场景中的位置信息作为PHDF的输入.SMC-PHD滤波器对检测结果进行滤波,实现对观测场景中运动目标数量和目标状态的估计.传统SMC-PHDF由于不对目标进行标记避免了数据关联,但同时也丧失了对单个目标航迹进行持续跟踪的能力.为此,提出采用标记粒子及最近邻聚类构建关联决策,根据粒子标记经重采样后的统计分布计算最大关联概率实现当前目标与航迹的时域关联.实验证明,当观测场景中的目标数目、状态随时间变化且检测结果存在虚警情况下,该算法能较好地估计多目标数量和状态,其时域关联准确性比MHT算法更高.%A multiple visual targets tracking algorithm based on sequential Monte Carlo probability hypothesis density filter (SMC-PHDF) is proposed. Background modeling and dynamic object detection was implemented on observed scenes via algorithm of W4. Positions of possible objects were gotten as the input of PHDF. SMC-PHDF had the detection results filtered, and achieved estimates of object number and state. Traditional SMC-PHDF, which does not label objects to avoid data association , is unable to continuously track individual object. Labeled particles as well as nearest clustering are proposed to construct association strategy, and the association of current objects and former trajectories can be realized via inheriting and propagation of particles. Experiments verify that this algorithm can get good estimates of object number and state when target number as well as states changes and clutters exist. Its accuracy on association is higher than that of MHT ( multiple hypotheses tracking).
展开▼
