【摘要】 提出一种基于Mean Shift改进算法与连通域标记的多目标跟踪算法。在多目标跟踪过程中,对目标瞬间丢失、目标遮挡或重叠时目标跟踪失败等情况有较好的改进。在跟踪过程中,当目标丢失时,基于改进的Mean Shift算法能自适应调整搜索窗口尺寸和方向。通过自适应扩展搜索窗口,利用连通域标记算法搜索目标并计算其矩特征来获得跟踪目标的重心和大小信息,并将获得的位置和尺度信息作为下一帧Mean Shift算法跟踪初始坐标和尺度,统计目标区域颜色直方图作为Mean Shift算法目标模型,从而解决了因目标速度过快而引起的目标瞬间丢失问题。最后研究结果显示,这种改进的目标跟踪算法可以有效改善多目标跟踪的性能,实现目标连续跟踪。更多还原

【Abstract】 This paper proposed an improved Mean Shift based algorithm and a connected components labeling algorithm for multi-target tracking,which had a good performance in the process of multi-target tracking to deal with the situations as target instant loss,target occlusion and target tracking failure during the overlapping.When the target was missing in the tracking,this improved tracking algorithm—Mean Shift could be self-adapted to adjust the size and direction of search window.What’s more,through the self-adaptive extended search window,it made use of connected components labeling algorithm to search target and calculate its moment character as to acquire the tracking target focus and size,and the getting target focus and size will be the tracking initial coordinate and size of Mean Shift in the next frame.Calculated the color histogram of the object region and put the histogram as the object model.That could solve problems as target loss resulted from the over-speed target moving.The final results elicit that this kind of improved target tracking algorithm can be efficient in ameliorating the function of multi-target tracking and realizing target’s continuing tracking.更多还原