基于Mean shift的核窗宽自适应目标跟踪新算法  

New Tracking Algorithm Based on Mean shift with Adaptive Bandwidth of Kernel Function

作　　者：王勇[1,2] 谭毅华[1] 田金文[1] 

机构地区：[1]华中科技大学多谱信息处理技术国防重点实验室,武汉430074 [2]中国地质大学机电学院,武汉430074

出　　处：《数据采集与处理》2009年第6期762-766,共5页Journal of Data Acquisition and Processing

摘　　要：针对传统均值漂移算法(Mean shift)中核函数直方图对目标特征描述较弱、跟踪过程中核函数带宽的保持不变的缺点,提出了一种新的核函数带宽可变的Mean shift跟踪算法。在特定的色彩空间中,统计落入各区间的像素个数,并对各区间像素的位置建立高斯分布模型,采用二阶空间直方图实现目标建模,强化目标特征描述提高了跟踪的鲁棒性;结合边缘检测与角点检测选取目标特征点估算目标仿射模型确定伸缩尺度,适应目标多自由度变化下的跟踪。实验结果证明,该算法比原有算法跟踪效果更加准确和可靠。The traditional Mean shift is invariable bandwidth and cannot represent accurately the color distribution of the object. To improve theoretical limitation, a novel target tracking algorithm within the framework of Mean shift is presented. Firstly, a new color space of the object is partitioned into subspaces by considering the weighted number of pixels with feature vectors cluster, and describing the pixel coordinates with Gaussian distribution. Then, an im- proved spatiograms is adopted to describe the reference model and the candidate model, and Bhattacharyya distance is derived to evaluate the similarity between them. Finally, the affine transform is established by combining corner and edge detector to calculate scale parameter. The algorithm is proved to have better effect and robust through experiments.

关 键 词：目标跟踪 均值漂移算法 空间直方图 可变带宽 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN911.73[自动化与计算机技术—计算机科学与技术]