基于子图像的尺度自适应Mean shift目标跟踪

Scale adaptable mean shift object tracking based on image fragments

针对经典Mean shift目标跟踪算法在目标被长期遮挡情况下容易产生跟踪误差甚至跟丢目标,提出了一种改进的Mean shift目标跟踪方法.该方法将被跟踪目标划分为多个图像子块,通过对多个子图像赋予不同的权重融入目标的空间信息,目标模板与候选目标之间的相似性系数由对应的多个子图像的Bhattacharyya相似性系数融合而成.实验结果表明,该方法对被长期遮挡的目标能进行稳健、高效的跟踪.在传统尺度自适应策略的基础上利用边缘直方图方法,通过当前帧与初始帧中目标边缘直方图的Bhattacharyya相似性系数来进一步判断目标尺度是否真的减小.实验结果表明,该算法能很好地实现尺度变化.

The classical mean shift tracking algorithm is apt to make errors or lose the target if the target is occluded for a very long time. Thus an improved mean shift tracking algorithm is proposed. This algorithm divides the target into multiple fragments and integrates spatial information by using different weights of each image fragment. The similarity coefficient between target template and candidate template consists of the Bhattacharyya coefficients of the corresponding multiple fragments. Experimental results show that the proposed method is efficient when the target is occluded for a long time. A new method named edge-histogram is used. This method is based on original scale updating mechanism and make a further judgment that whether the target is smaller or not by calculating the Bhattacharyya coefficient between the target's edge-histograms of the current frame and the previous one. Experimental results show that the proposed algorithm can deal with the scale problem very well.