对步态空时数据的连续特征子空间分析  

Recursive spatiotemporal subspace learning for gait recognition

作　　者：胡荣[1] 王宏远[1] 

机构地区：[1]华中科技大学电信系数字视频与通信中心,武汉430074

出　　处：《中国图象图形学报》2011年第4期625-633,共9页Journal of Image and Graphics

基　　金：高等学校科技创新工程重大项目培育资金资助项目(708065)

摘　　要：提出一种基于空时特征提取的人体步态识别算法。连续的特征子空间学习依次提取出步态的时间与空间特征:第一次特征子空间学习对步态的频域数据进行主成分分析,步态数据被转化为周期特征矢量;第二次特征子空间学习对步态数据的周期特征矢量形式进行主成分分析加线性判别分析的联合分析,步态数据被进一步转化为步态特征矢量。步态特征矢量同时包含运动的周期特征以及人体的形态特征,具有很强的识别能力。在USF步态数据库上的实验结果显示,该算法识别率较其他同类算法有明显提升。A gait recognition method based on spatiotemporal feature extraction is proposed. Recursive subspace learning is used to extract both time and space feature of gait. In the first subspace learning, the periodic dynamic feature of gait is extracted by principal component analysis and sequence data is represented in the periodicity feature vector form. In the second subspace learning, principal component analysis plus linear discriminant analysis are applied to the oeriodieity feature vector representation of gait and sequence data is compressed into gait feature vector, gait feature vector is an effective representation because it contains both human dynamic and shape feature. Experimental result on the USF gait database shows that the proposed method achieves highly competitive performance with respect to other published gait recognition approaches.

关 键 词：步态识别 子空间学习 主成分分析 线性判别分析 

分 类 号：TP301.2[自动化与计算机技术—计算机系统结构]