Skeleton Silhouette Based Disentangled Feature Extraction Network for Invariant Gait Recognition

Abstract

Recently, the demand for biometric recognition systems through deep-learning algorithms has increased, and related researches have been actively underway. Gait has its unique characteristics like fingerprints and is receiving much attention due to its capability to obtain from a long distance without contact. Gait recognition means to identify an individual by analyzing a person’s gait patterns, and can be applied to various fields such as criminal investigations. However, the accuracy of gait recognition suffers from external factors such as variation of view angles and clothes. To address this problem, we propose a skeleton silhouette-based disentangled gait recognition network, which learns view-invariant features. Experimental results on CASIA-B dataset are also shown comparing with previous methods.