计算机工程 ›› 2013, Vol. 39 ›› Issue (8): 181-186.doi: 10.3969/j.issn.1000-3428.2013.08.039

• 人工智能及识别技术 • 上一篇    下一篇

机器人的人体姿态动作识别与模仿算法

李少波1,赵毅夫2,赵群飞1,阎威武1   

  1. (1. 上海交通大学自动化系系统控制与信息处理教育部重点实验室,上海 200240; 2. 复旦大学计算机科学技术学院,上海 201203)
  • 收稿日期:2012-08-31 出版日期:2013-08-15 发布日期:2013-08-13
  • 作者简介:李少波(1987-),男,硕士研究生,主研方向:图像处理,模式识别;赵毅夫,本科生;赵群飞,教授、博士;阎威武,副教授、博士
  • 基金项目:
    国家自然科学基金资助项目(60974119)

Algorithm of Human Posture Action Recognition and Imitation for Robots

LI Shao-bo 1, ZHAO Yi-fu 2, ZHAO Qun-fei 1, YAN Wei-wu 1   

  1. (1. Key Laboratory of System Control and Information Processing, Ministry of Education, Department of Automation, Shanghai Jiaotong University, Shanghai 200240, China; 2. School of Computer Science, Fudan University, Shanghai 201203, China)
  • Received:2012-08-31 Online:2013-08-15 Published:2013-08-13

摘要: 提出从人体姿态估计的角度实现人与NAO机器人的互动交流。用AdaBoost算法训练得到人体部件检测器,利用得到的部件检测器在图片中寻找到对应部件的若干候选区域,在寻找的过程中兼顾部件的位置和方向。依据提出的评价指标评价这些候选区域,从中选出最优。以NAO机器人为实验平台,在其动作可达范围内实现对人体动作姿态的识别与模仿,达到较好的动作空间一致性和时间一致性。实验结果证明该算法可有效识别和模仿人体姿态。

关键词: 人机交互, NAO机器人, 人体姿态估计, AdaBoost算法, 部件检测器, 人体姿态模仿

Abstract: This paper takes a view of human pose and action recognition to realize human-robot interaction. A part detector for each human body part is obtained by AdaBoost algorithm. With these detectors, several candidate regions for the corresponding body part are found. During the finding process, the position and orientation of parts are both considered. These candidate regions are evaluated by using several defined evaluation indexes in order to select an optimal one. Human pose recognition and action imitation are implemented by using a NAO robot as test bench, and NAO’s performance shows good spatial consistence and temporal consistence within its ability of operation. The result shows that the algorithm can effectively identify and imitate human posture.

Key words: human-robot interaction, NAO robot, human pose recognition, AdaBoost algorithm, part detector, human posture imitation

中图分类号: