为解决Mean-shift算法采用固定跟踪窗口造成的目标定位精度低的问题, 结合视觉显著性检测和像素灰度相似度, 提出一种采用自适应核函数的Mean-shift跟踪算法。该方法以灰度相似度加权的视觉显著性特征确定目标区域, 并结合Epanechnikov核函数构建自适应核函数, 使跟踪窗口自适应目标大小变化, 降低目标尺度变化的影响, 实现目标的有效跟踪。实验结果证明, 该方法能够有效跟踪尺度变化目标, 处理每帧图像耗时小于25 ms, 满足实时性需求。

In order to solve the problem of Mean-shift algorithm caused by the fixed track window, an improved Mean-shift algorithm using adaptive kernel function is proposed. Visual saliency weighted by the gray similarity is detected to ascertain the object area, and the adaptive kernel function is designed to track object combined with Epanechnikov and the object area, reducing the effect of fixed track window and background pixels. After plenty of experiments, the results show that the proposed method can track object scale motions in real time and exactly, and cost less than 25 ms for every frame.