An Invariant Appearance Model for Gait Recognition

This paper describes a novel feature representation method for gait analysis and recognition applications. By tiling one-period gait subsequence in a 2D polar-plane along a ring frame by frame, a gait appearance model is built. The model consists of structural information of individual silhouette and contextual silhouettes centered at the current frame in the polar-plane. With an invariant histogram-based descriptor, gait appearance characteristic is described as a sequence of shape distributions. These distributions are employed to achieve gait recognition based on Jeffrey divergence criterion and dynamic time warping. Furthermore, the effect of viewing angle to gait period estimation and recognition performance is explored. Experimental results on CASIA database demonstrate that the proposed approach has a better performance than existing methods.