Dynamic texture based gait recognition

Abstract

We present a novel approach for human gait recognition that inherently combines appearance and motion. Dynamic texture descriptors, Local Binary Patterns from Three Orthogonal Planes (LBP-TOP), are used to describe human gait in a spatiotemporal way. We also propose a new coding of multire-solution uniform Local Binary Patterns and use it in the construction of spatiotemporal LBP histograms. We show the suitability of the representation for gait recognition and test our method on a popular CMU MoBo dataset. We then compare our result to the state of the art methods. © Springer-Verlag Berlin Heidelberg 2009.