Human gait represents an attractive biometric modality to re-identify a person as it requires non contact and it is perceivable at a distance. However, the view angle variation and the presence of covariate factors cause significant difficulties for recognizing gaits. In order to deal with such constraints, this paper presents a Part View Transformation Model (PVTM) for gait based applications. Compared with previous methods, the PVTM is applied on selected relevant parts chosen through a semantic classification step. Conducted on the CASIA-B gait database, experimental results show that the proposed method outperforms well known multi-view methods even under covariate factors (i.e. carrying bag, clothing).

人的步态代表了一种有吸引力的生物特征识别方式，可以重新识别一个人，因为它不需要接触并且可以在远处感知到。然而，视角变化和协变量因素的存在给识别步态带来了很大的困难。为了解决这些限制，本文提出了一种针对基于步态的应用的零件视图转换模型（PVTM）。与以前的方法相比，PVTM应用于通过语义分类步骤选择的相关部分。进行的CASIA-B步态数据库上，实验结果表明，所提出的方法优于公知的，即使在协变量因素多视图的方法（即携带包，衣服）。