一种拥挤环境下的多目标跟踪算法

Multiple targets tracking in a crowded environment

多目标的鲁棒跟踪是视频监控系统的基础。在拥挤的环境下,由于遮挡的原因,传统的单目跟踪方法很难分割前景目标并跟踪。本文通过码本算法获得多个视角的前景信息,利用部分标定法获得平面单应性矩阵,根据此矩阵将各个视角的前景信息投射到参考视角后进行数据融合,利用得到的定位信息进行跟踪。实验结果表明,该算法在拥挤的环境中能实现对多目标的鲁棒跟踪,具有很好的实时性。

Robust multi-target tracking is the basis of video surveillance systems.In a crowded environment,it is difficult for traditional monocular method to segment foreground targets and track.The foreground information through codebook algorithm and the homography matrix are obtained by partial calibration method,according to this matrix,the foreground information of each view is projected into the reference view to make data fusion and track by the localization information.Experimental results show that this algorithm can achieve robust tracking of multiple targets in a crowded environment and works real-time.