Cloth Changing Person Re-Identification Based on Backtracking Mechanism

Abstract

Person Re-Identification (Re-ID) plays a vital role in intelligent surveillance systems, enabling the tracking of individuals across different cameras. However, most existing systems rely heavily on clothing-based features, leading to performance failures when a person changes attire. To overcome this limitation, this paper presents a novel approach Cloth-Changing Person Re-Identification based on the Backtracking Mechanism. The system combines identity and body-shape features to achieve clothing-invariant recognition. A dual-branch architecture utilizing ResNet50 for appearance-based feature extraction and HRNet for body-shape modeling is implemented. The proposed feature infiltration and clothes suppression loss functions enhance discrimination while minimizing clothing bias. The approach achieves improved recognition accuracy under varying apparel conditions, making it practical for long-term surveillance and public safety. This integrated framework strengthens intelligent monitoring systems, ensuring reliable identification despite clothing variations and contributing significantly to public safety and security advancements.