万向椭圆描述的Mean-Shift算法

Mean-Shift Algorithm Described by Irregular Ellipse

传统的Mean-Shift算法在目标跟踪过程中,由于跟踪窗口尺度固定而不能很好适应目标的尺度变化,当目标尺度减小时,目标区域所提取的特征向量包含过多的背景干扰信息,目标尺度增大会使跟踪窗口偏离目标的质心,降低跟踪的鲁棒性。为此文中采用万向椭圆的方式对目标区域进行描述,减少背景干扰信息以突出目标模型,提取椭圆区域的加权颜色直方图为目标特征,采用尺度加减法自适应调整椭圆区域的大小,并在跟踪过程中根据运动轨迹动态调整椭圆方向,以增强跟踪的准确性。实验结果表明万向椭圆能够更好地描述跟踪目标的尺度和方向,在目标尺度变化比较平稳的情况下,尺度加减法能自适应调整跟踪窗口的尺度,可以取得良好的跟踪效果。

During the process of the traditional Mean-Shift algorithm for object tracking,the size of tracking window is fixed and cannot adapt to the change of target scale. When the target scale decreases,the extracted feature vectors of the target region contains too much background interference information,the increasing of target scale deviates tracking window from the target will reduce the robust of tracking. So in this paper,use irregular ellipses to describe the outline of target instead of rectangle,reduing the background interference information in order to highlight the target model,extracting weighted color histgram for the target characteristics of elliptical area. It uses the addition and subtraction for adapting to the size of elliptical areas and in the process of tracking based on trajectory dynamic adjusts to elliptical direction to enhance the tracking accuracy. The experimental results show that the universal elliptic can better describe the target tracking scale and direction,in the case that target scale variations are stable,addition and subtraction can adaptively adjust to the tracking window scale,can achieve good tracking effects.