基于多层定位的多目标跟踪算法

在遮挡频繁发生的环境中,多目标跟踪是一个难题.为解决此难题,提出了一种基于多层定位的多目标跟踪算法.该算法利用码本模型检测前景.首先计算垂直方向上的灭点,进而获得基于多层的单应性矩阵,完成多目标在多层上的定位.然后利用图割算法进行跟踪.该算法不需要相机的完全标定,提高了计算效率.实验结果表明,本算法实时性好,对多目标之间的遮挡具有很强的鲁棒性.

Tracking algorithm of multi-object based on localizing at multiple planes

When occlusion between multiple objects is common,tracking of multiple objects is a difficult problem.In order to deal with this problem,a novel algorithm to localize and track multi-object on multiple planes is proposed.The foreground is detected by using codebook model.Then,the vanishing points of the vertical direction are computed and this information is used to compute homography matrix induced by multiple planes.Multiple objects are located at multiple planes.Finally,tracking is performed by graph cuts.The algorithm does not require fully camera calibration,and the complexity of the algorithm is reduced.The experimental results show that multiple objects can be tracked reliably and the algorithm has nice robustness to the occlusion in crowed environments.