基于多信息融合的Mean-Shift跟踪算法

Mean-Shift tracking algorithm based on multi-information fusion

针对Mean-Shift跟踪算法存在的问题,提出一种以Mean-Shift跟踪器为框架,结合在线学习检测器以及Kalman滤波器的动态目标跟踪算法。利用Mean-Shift跟踪器跟踪目标物,引入在线学习检测器可以在跟踪器跟踪目标失败后,快速检测目标并更新目标模板,通过引入Kalman滤波器能够提高跟踪精度以及检测器的检测速度,最后将三者信息融合,实现对目标物长时间有效跟踪。实验结果中目标重叠度和每秒传输帧数分别在0.70和18帧/s以上,表明:所提算法具有跟踪精度高、实时性强等优点,能够有效解决光照变化,部分遮挡或全部遮挡等导致目标物跟踪丢失的问题。

Aiming at the problems existing in the Mean-Shift tracking algorithm,a dynamic target tracking algorithm based on Mean-Shift tracker and the online learning detector and Kalman filter is proposed.First use the Mean-Shift tracker to track target,after the tracker fails,online learning detector can be used to detect targets and timely update target template,so as to ensure the reliability of the tracker. The introduced Kalman filter not only improves the tracking precision,but also estimates the target area to improve the detecting speed of the detector.The effective integration of the three above can effectively solve the light changes,partial occlusion or all occlusion and others which cause problem of target tracking loss.The target overlap and the frames transmitted per second are above 0.70 and 18 fps,which show that the proposed algorithm has the advantages of high tracking precision and strong real-time performance.