改进联合概率数据关联的视频多目标快速跟踪

针对监控范围较大、目标外观特征少的视频多目标数据关联及跟踪问题，本文仅利用目标运动特征，提出了一种基于联合概率数据关联（joint probabilistic data association，JPDA）的复杂情况下视频多目标快速跟踪方法．首先采用murty算法求JPDA的最优K个联合事件，大大降低了计算复杂度；然后根据JPDA的关联概率讨论目标的运动情况，分析在多目标新出现、遮挡、消失、分离（前景检测存在目标碎片）等复杂情况下当前帧量测与跟踪目标的数据关联问题，获取复杂运动的多目标跟踪轨迹．在多个监控视频上的实验结果表明，该方法能大大提高跟踪性能，实现复杂情况下的视频多目标快速跟踪．

Tracking multiple video objects based on improved joint probabilistic data association

For the data association of video objects having little distinguishable features in large-scale monitoring scenes, we present a method for tracking multiple video objects in real-time based on the joint probabilistic data association (JPDA), in which the motion features of the objects are incorporated. First, the k-best joint events are computed by the Murty algorithm to reduce the complexity, and then, the motion situations of objects are analyzed by the association probability of JPDA. When objects are entering and exiting the field of view, merging and splitting (objects are detected as fragmented parts), the data association algorithm acquires the tracking trajectories of the objects. Experiments demonstrate the feasibility and performances of the proposed approach.