华南理工大学学报(自然科学版) ›› 2015, Vol. 43 ›› Issue (1): 99-104.doi: 10.3969/j.issn.1000-565X.2015.01.016

• 计算机科学与技术 • 上一篇    下一篇

面向真实监控环境的前视步态识别方法

姜颖军 王建新 郭克华   

  1. 中南大学 信息科学与工程学院, 湖南 长沙 410083
  • 收稿日期:2014-05-19 修回日期:2014-08-03 出版日期:2015-01-25 发布日期:2014-12-01
  • 通信作者: 姜颖军(1972-),男,博士,高级工程师,主要从事机器学习研究 . E-mail:124601037@csu.edu.cn
  • 作者简介:姜颖军(1972-),男,博士,高级工程师,主要从事机器学习研究 .
  • 基金资助:
    国家自然科学基金青年科学基金资助项目( 61202341/F020508 );国家自然科学基金重点资助项目( 61232001/F02 )

A Novel Front-view Gait Recognition Approach in Real Surveillance Environments

Jiang Ying-jun Wang Jian-xin Guo Ke-hua   

  1.  School of Information Science and Engineering , Central South University , Changsha 410083 , Hunan , China 
  • Received:2014-05-19 Revised:2014-08-03 Online:2015-01-25 Published:2014-12-01
  • Contact: 姜颖军(1972-),男,博士,高级工程师,主要从事机器学习研究 . E-mail:124601037@csu.edu.cn
  • About author:姜颖军(1972-),男,博士,高级工程师,主要从事机器学习研究 .
  • Supported by:
    Supported by the National Natural Science Foundation for Young Scholars of China ( 61202341/F020508 ) and the State Key Program of National Natural Science Foundation of China ( 61232001/F02 )

摘要: 真实监控环境下,行人前视步态比侧视步态更常见 . 现有步态识别方法主要针对侧视步态而非前视步态 . 为此,文中根据行人步态统计特征,提出了一种基于自动视角转换的前视步态识别方法 . 该方法通过计算行人步态能量图、行走迹线和步态视角,提取经视角转换后同一视角下的步态特征并进行比对识别 . 实验结果表明,在真实监控环境且单目监控摄像机参数未知的情况下,该方法对前视步态的正确识别率达 81% ,每秒可识别 21 帧,具有良好的识别效果 .

关键词: 步态识别, 生物识别, 步态能量图, 视角不变性, 特征提取

Abstract: In real monitoring circumstances , pedestrians ’ front-view gaits are more common than lateral-view ones. As the existing gait recognition methods are mainly focused on lateral-view gaits instead of front-view gaits ,a new front-view gait recognition method on the basis of automatic perspective conversion is proposed according
to the statistics of pedestrians ’ gait features. Through calculating pedestrians ’ gait energy images , walking
trajectories and gait view angles , this method extracts and recognizes gait features under the same view angle
after the transformation of view angles. Experimental results show that the proposed method works well under
real monitoring circumstances when the parameters of monocular camera are unknown , and that it achieves a
recognition rate for front-view gait of 81% and an identified frame number of 21 per second.

Key words: gait recognition, biological recognition, gait energy image, view invariant, feature extraction