Abstract: The Auto-Camshift method based on NMI feature and Camshift algorithm is put forward. This method segments the images, then utilizes Camshift algorithm to process each object and NMI feature to recognize objects. Compared with Camshift algorithm, this method adds NMI feature, which is the invariant under rotation, scaling, translation to recognize the objects. It can both identify multi objects at the same time and study automatically. Experiment results on Friend-Robot system indicate that the algorithm is effective and practical to meet the real time needs for autonomous robot vision.

Key words: Camshift; NMI feature; Robot vision; Auto-Camshift

摘要： 基于NMI 特征及Camshift 算法，提出了扩展Auto-Camshift 算法，该算法分割得到目标物体，分别对每个物体进行Camshift 运算，同时结合了物体归一化转动惯量特征对物体进行学习和辨认。该算法同Camshift 相比，增加了归一化转动惯量NMI 这一以目标不变特征为识别基础的特征，同时实现了多物体识别、自主学习。在“富莱德”机器人系统上的试验结果证明所提供的算法在学习和识别系统中有很好的鲁棒性和实时性。

关键词: Camshift；NMI 特征；机器人视觉；Auto-Camshift