基于改进直方图映射和均值移位目标跟踪算法

Object Tracking Using Improved Histogram Back-projection and Mean-shift

提出了一种改进直方图映射和均值移位结合的目标跟踪算法,根据目标模型各成分直方图比例系数建立索引表,索引号为颜色值,索引内容为颜色对应的归一化目标模型直方图,在索引表中查询输入图像颜色值构建反映目标出现概率灰度图;均值移位算法在生成灰度图中快速定位目标位置。对比实验结果表明,改进的直方图映射算法从根本上抑止模型背景干扰,均值移位算法准确定位目标;系统能实时准确跟踪运动目标。

This problem is relieved in some measure, but not solved fundamentally. An improved back-projection is proposed in this paper that can suppress background interference of object model fundamentally. A normalized histogram is produced in object model. An indexed list is built, the index variable is the gray value and the index content is the histogram value. A back-projection gray image is created by searching the indexed list by the input image color value. Mean-shift finds the local maxima to locate the object near the neighborhood of object center in the previous frame. Experimental results show the algorithm can track the moving object precisely in real-time.