Long-term person re-identification for access control using continuous learning of clothing information

Abstract

Camera-based unaware access control is difficult because of posture and clothing changes. Long-term Re-ID is a task to cope with such posture and clothing changes. This study focuses on the application of Long-term Re-ID to access control and the possibility of improving the existing person identification accuracy and robustness by continuously learning clothing. The goal is to extend the continuous learning task to a person re-identification task in order to continuously utilize clothing information to improve the accuracy and robustness of known person identification.