计算机与现代化 ›› 2012, Vol. 1 ›› Issue (200): 192-04.doi: 10.3969/j.issn.1006-2475.2012.04.052

• 图像处理 • 上一篇    下一篇

基于单一深度图像的人体姿态实时识别技术研究

杜霄鹏1, 郝建平1, 李星新1,杨 俊2   

  1. 1.军械工程学院装备指挥与管理系,河北 石家庄 050003; 2.中国人民解放军68210部队司令部,陕西 宝鸡 721000
  • 收稿日期:2011-10-31 修回日期:1900-01-01 出版日期:2012-04-16 发布日期:2012-04-16

Human Pose Recognition Research Based on Single Depth Images

DU Xiao-peng1, HAO Jian-ping1, LI Xing-xin1, YANG Jun2   

  1. 1. Department of Equipment Command and Management, Ordnance Engineering College, Shijiazhuang 050003, China;2.Headquarters of Unit 68210, Chinese People’s Liberation Army, Baoji 721000, China
  • Received:2011-10-31 Revised:1900-01-01 Online:2012-04-16 Published:2012-04-16

摘要: 为探索更自然、逼真的交互方式,对基于深度图像技术进行研究。介绍当前深度图像技术的应用现状以及主要研究方法;利用深度图像对人体进行识别,包括基于多幅深度图像和基于单一深度图像;对于人体姿态静态追踪,基于当前研究成果将人体部位进行分割处理,以计算机处理速率及鲁棒性为出发点,将随机森林算法应用于单一深度图像中人的定位,并提出改进方法,实现身体部位的识别以及骨骼关节点的空间位置精确标定。通过试验分析对人体深度图像识别速率及精确度方面的改进效果进行验证。

关键词: 深度图像, 姿态识别, 随机森林, Mean Shift方法

Abstract: To obtain more natural and realistic interactive mode, the research based on depth image technology is carried on. The situation of depth image application and methods are introduced. The depth image is used to recognize human pose induding based on multiple depth image or single depth image. And for the static human pose tracking, random forest algorithm is used to evaluate the human position in the single depth image for improving the run rate and the robust character. Then the recognition of body position and the skeleton joint are realized. At last, an instance is provided to illustrate the effectiveness and accuracy of the method.

Key words: depth image, human pose recognition, random forest, Mean Shift method