摘要： 传统Harris检测算法不能很好地适应跟踪环境。为此，提出一种基于角点特征融合的Mean-shift跟踪算法。考虑人体姿态变化或遮挡对多区域跟踪的影响，采用角点更新策略，将特征融合主色调模型的跟踪结果与多区域跟踪结果进行权衡。实验结果表明，该算法能克服人体姿态变化或遮挡对跟踪的影响，实时性满足一般跟踪系统的要求，且在非遮挡状况下，其跟踪准确率比传统算法高。

关键词: Mean-shift算法, 角点检测, 主色调, 多区域, 部分遮挡, 姿态变化

Abstract: For traditional Harris detection algorithm cannot suit the situation of tracking well, a algorithm of corner point detection which depends on different ellipse regions is proposed. For the influence of body attitudes variation or occlusion, a strategy of points updating is adopted simultaneously. The result of multi-zone tracking against the global tracking based on dominant hue and multi-features fusion is balanced. Experiments show the algorithm proposed can overcome the body posture change or occlusion effectively, meets the requirement of real-time, and the accuracy of tracking in non-occluded environment is also higher than traditional algorithm.

Key words: Mean-shift algorithm, corner detection, dominant hue, multi-zone, partial occlusion, attitude variation

中图分类号: 

• TP391