联合Kalman和自适应Mean-Shift稳健相关视频跟踪方法

Joint Kalman and Adaptive Mean-Shift Based Robust Correlative Visual Tracking Algorithm

相关视频跟踪器存在计算量大、模板漂移、对机动目标,杂波影响大以及遮挡情况无法跟踪的问题,而Kalman滤波能通过利用相关跟踪器的输出结果来预测目标在下一帧里在图像中的坐标,可以在高概率的小范围内对目标进行搜索,以大幅减小计算量和杂波的影响。然后,当跟踪器由于受到杂波或遮挡的影响而提供了错误的测量信息时,跟踪的性能将大幅下降。大量研究表明,Mean-Shift跟踪器具有运算速度快和跟踪性能好的特点,而当目标柱状图和待选图像区域相近时,其跟踪性能也将大幅下降,甚至无法进行跟踪。为了解决该问题,结合上述3种思想提出了一种改进的、稳健的视频目标跟踪方法,并通过理论分析和仿真结果表明了算法的有效性和优越性。

Correlation tracker has the problem of computation intensive （if the search space or the template is large）, template drift and may fail in case of fast maneuvering target, occlusion suffered by it and clutter in the scene. By using the output resuhs of correlation tracker, Kalman fiher can predict the larget coordinates in the next frame. Thus, the target may be searched within a small range with a high probability. In this way, the amount of calculation and the influence of clutter can be sharply reduced. However, if wrong measurement vector is provided to the tracker due to the clutter or the occlusion inside the search region, the performance of tracking will fell sharply. Fast operation speed and good tracking results has shown to Mean-shift tracker in the literature, but it may fail when the histograms of the target anti the candidate region in the scene are similar . In order to deal with the mentioned problems, an improved robust visual target tracking method based on the three above ideas is proposed, and the algorithm is showed effectively and superiorly through the theoretical analysis and simulation resuhs.