Teepe Torben ; Khan Ali ; Gilg Johannes ; Herzog Fabian ; Hörmann Stefan ; Rigoll Gerhard

Abstract

Gait recognition is a promising video-based biometric for identifyingindividual walking patterns from a long distance. At present, most gaitrecognition methods use silhouette images to represent a person in each frame.However, silhouette images can lose fine-grained spatial information, and mostpapers do not regard how to obtain these silhouettes in complex scenes.Furthermore, silhouette images contain not only gait features but also othervisual clues that can be recognized. Hence these approaches can not beconsidered as strict gait recognition. We leverage recent advances in human pose estimation to estimate robustskeleton poses directly from RGB images to bring back model-based gaitrecognition with a cleaner representation of gait. Thus, we propose GaitGraphthat combines skeleton poses with Graph Convolutional Network (GCN) to obtain amodern model-based approach for gait recognition. The main advantages are acleaner, more elegant extraction of the gait features and the ability toincorporate powerful spatio-temporal modeling using GCN. Experiments on thepopular CASIA-B gait dataset show that our method archives state-of-the-artperformance in model-based gait recognition. The code and models are publicly available.