应用Mean Shift和分块的抗遮挡跟踪

针对传统Mean Shift跟踪算法在目标发生遮挡时容易跟偏甚至跟丢的缺陷,提出了一种新的抗遮挡跟踪算法。首先,对跟踪窗口内的目标进行分块;然后,对外围子块分别实施Mean Shift跟踪算法并检测遮挡的发生,当遮挡发生后即对所有子块实施Mean Shift跟踪算法;最后,引入一种子块置信度机制并仅用置信度最高的子块来确定目标的最终位置,从而在目标发生遮挡时能有效剔除被遮挡子块对目标定位的影响。对不同的视频序列测试的结果显示,本算法能对发生遮挡的目标进行准确跟踪。当遮挡目标尺寸为70pixel×100pixel时,平均处理时间为38.6ms/frame。结果表明,改进算法能够满足目标跟踪系统稳定性和实时性的要求。

Anti-occlusion tracking algorithm based on Mean Shift and fragments

A new anti-occlusion tracking algorithm is presented to solve the problem that traditional Mean Shift based tracking algorithm often deviates or loses the targets under occlusions. Firstly, the target in the tracking window is divided into a number of fragments,then the Mean Shift algorithm is used to move the peripheral fragments separately and to detect the occlusions.Furthermore, all fragments can be computed when the target is occulated. Finally, the confidence of each fragment is computed and only the fragment with the highest confidence is involved to achieve the whole target’s coordinates and to avoid the influence of occluded fragments on the target location. Tested results for different video sequences indicate that proposed algorithm can track precisely the target occluded. When the occlusion size is 70 pixel×100 pixel, the running time is 38.6 ms/frame,which meets the requirements of target tracking system for the stability and real time.