摘要
针对在提取步态轮廓特征时,步态识别算法复杂、运算时间长、难以满足实时性需求的问题,提出了一种基于稀疏表示及分段帧差能量图的步态识别方法.首先,建立改进的分段帧差能量图(SFDEI)作为步态的特征图像;对每个分段的帧差能量图建立字典,采用改进的正交匹配追踪算法对系数快速分解;最后,应用隐马尔可夫模型(HMM)对改进的分段帧差能量图建立步态识别模型.实验采用CASIA B步态数据库,以90视角进行实验.结果表明方法有较高识别率,同时可满足实时性需求.
In order to deal with the problems that the gait recognition algorithms are complicated, cost a lot of time and can＇t be used in real-time condition in extracting the feature of the gait silhouette, a method based on sparse representation and segmented frame difference energy image is proposed. Firstly, the improved segmented frame difference energy image is established as a feature image of gait. Then a dictionary is established based on it and an improved orthogonal matching pursuit algorithm is used for the fast coefficient decomposition. Finally, hidden Markov model is applied to establishing the gait recognition model based on the improved partial frame difference energy image. Selecting 90° views, an experient is done on CASIA_B database. Experimental results show that this methods get better recognition rate and can be used in real-time system.
出处
《信息与控制》
CSCD
北大核心
2013年第1期27-32,共6页
Information and Control
基金
国家自然科学基金资助项目(61005032)
关键词
稀疏表示
隐马尔可夫模型
正交匹配追踪算法
分段帧差能量图
步态识别
sparse representation
hidden Markov model （HMM）
orthogonal matching pursuit
segmented frame difference energy image
gait recognition
