Gait recognition using dynamic gait energy and PCA+LPP method

In this paper we propose a gait recognition method with dynamic gait energy image (DGEI) and manifold learning. First we present a new gait feature-dynamic gait energy image which can reflect the dynamic variance parts of the motion body and can better characterize gait features. Secondly in order to preserve the principal and discriminant components, we use PCA and LPP to discover the low-dimensional manifold of the high feature space, in which the characteristics of DGEI are well preserved. Lastly the simple vote rule and Dempster-Shafer (D-S) evidential theory are used as the fusion strategy for fusing multi-views gait information, the experimental results show D-S fusion method can get better recognition performance.