计算机工程与应用 ›› 2013, Vol. 49 ›› Issue (18): 134-137.

• 图形图像处理 • 上一篇    下一篇

基于关键帧的人体行为识别方法

成  勋1,常  飞2,吴志杰3   

  1. 1.中国工程物理研究院 研究生部,四川 绵阳 621900
    2.中国工程物理研究院 计算机应用研究所,四川 绵阳 621900
    3.中国工程物理研究院 纪委监察局,四川 绵阳 621900
  • 出版日期:2013-09-15 发布日期:2013-09-13

Human behavior recognition based on key frame

CHENG Xun1, CHANG Fei2, WU Zhijie3   

  1. 1.Graduate Division of China Academy of Engineering Physics, Mianyang, Sichuan 621900, China
    2.Institute of Computer Science and Application of China Academy of Engineering Physics, Mianyang, Sichuan 621900, China
    3.Commission for Discipline Inspection and Supervision Bureau of China Academy of Engineering Physics, Mianyang, Sichuan 621900, China
  • Online:2013-09-15 Published:2013-09-13

摘要: 提出了一种基于关键帧的人体行为识别方法。利用关于弧长的方向函数对人体形状进行描述以及通过形状空间中两点间的测地线距离对形状间的相似性进行度量,提取视频序列中的关键帧,通过模板匹配的方法实现对行为的识别。对CASIA单人行为数据库和Weismann数据库进行测试,实验结果表明该方法具有较高的识别率。

关键词: 行为识别, 关键帧检测, 相似性度量

Abstract: This paper addresses an approach which uses the direction function on the arc length to describe the body shape and the geodesic distance between any two points in the shape spaces to measure the similarity of shapes. It extracts key frames in the video sequence and recognizes the human action based on the template matching approach. The experimental results on the CASIA single behavior database and the Weismann database show that the method has a high recognition rate.

Key words: action recognition, key-frame detection, similarity measure