Abstract: Considering the kernel function bandwidth of traditional Mean Shift tracking algorithm cannot be adaptively adjusted and is liable to be effected in similar color, a multi-feature adaptive scale Mean Shift tracking algorithm is proposed. The algorithm combines color and texture information to create feature model, which creates the target probability density distribution in optimal target location, and calculates the wide and high of density blob, thus, the target scale is adaptively adjusted. Target the object with ellipse, parameters is obtained by calculating moment of probability density. Experimental results show that the algorithm can effectively meet the goal of scaling, rotation and other complex motion, and not sensitive to lighting change and similar color.

Key words: target tracking, kernel function bandwidth, texture, probability density, moment

摘要： 在传统均值漂移跟踪算法中，其核函数带宽缺乏较好的自适应调整特性，且易受背景色干扰。为此，提出一种多特征带宽自适应目标跟踪算法。采用颜色和纹理信息创建特征模型，在最优目标位置区域投影，以生成概率密度分布图，通过计算获得目标密度块的长度和宽度，从而自适应调整核函数带宽，用椭圆锁定目标，椭圆形状参数由目标概率密度的矩运算获得。实验结果表明，该算法能够有效适应目标缩放、旋转等复杂运动，并能抵御一定光照变化及背景色干扰影响。

关键词: 目标跟踪, 核函数带宽, 纹理, 概率密度, 矩

CLC Number: 

• TP301.6