Gait Recognition Based on Modified Gait Energy Image

Abstract

Biometric systems allow us to identify individuals from distinctive biological traits. Gait recognition is a biometric technique used to recognize humans based on the style of their walk. However, model-free based gait recognition performance is often degraded by the presence of some covariate factors such as view, clothing and carrying variations. From these, it has been shown that the change in appearance is the covariant that most affect the recognition performance. To address such issues, we propose to use a feature representation that takes both dynamic and static regions of silhouettes. This way, more robustness against covariates and better discriminative performance are expected. The proposed method is evaluated on one of the largest datasets available under the variations of clothing and carrying conditions: CASIA gait database B. Results show that the proposed method achieves correct classification rate up to 90% and outperformed state-of-the-art methods.