Human gait recognition with 3-D wavelets and kernel based subspace projections

Gait recognition can be regarded as a problem of uniquely representing spatiotemporal surfaces associated with a person?s walking pattern in an efficient manner. In this paper, we describe the approach of using projections of such surfaces onto subspace spanned by appropriate axes using a single framework. Two new algorithms for gait recognition are presented which use projection on subspace of kernel induced higher dimensional spaces using PCA and Fisher?s LDA. Wavelet transform in 3D is used to reduce the complexity of the problem. The proposed methods have been applied to datasets containing subjects walking at three different viewing angles in outdoor environment. The results show high accuracy results for recognizing subjects from their gait, even with a thumbnail size (16×16×12) of the gait patterns.