双向时空连续性轨迹片段关联的目标跟踪方法

摘要

提出一种使用双向时空连续性关联轨迹片段的目标跟踪方法.首先对检测结果进行简单的帧间匹配关联,生成可靠的轨迹片段;然后对每个轨迹片段通过卡尔曼滤波以及有权重的均值法分别计算修正轨迹片段的速度、位置与颜色特征;最后通过计算轨迹片段之间的双向时空连续性迭代关联,找到最符合时空连续性的轨迹片段关联.实验证明本文方法可以有效解决目标间以及目标被背景遮挡问题,实现对目标的稳定跟踪.%An object tracking algorithm by associating tracklets with the best bidirectional spatio-temporal continuity was proposed.First, reliable tracklets were generated by a primitive frame-by-frame association; then tracklet's motion, position and color features were eomputed and refined by applying Kalman,filter and weighted mean method respectively; finally, the best spatio-temporal association of tracklets was achieved through an iterative association by computing spatio-temporal continuity between tracklets.Experimental results prove that multiple objects can be successfully tracked under occlusion both by other object and scene object.