Robust Person Re-Identification Approach with Deep Learning and Optimized Feature Extraction

Jian Ding, Linze Li, Rongchang Li, W. Cong, X. Tianyang, Xiaojun Wu

Abstract: Person re-identification (Re-ID) is a critical technology in video surveillance, essential for enhanced security and efficient monitoring in urban settings. Our research improves the accuracy and robustness of Re-ID systems through advanced deep learning techniques and optimized feature extraction. By integrating methodologies such as non-local blocks, random erasing augmentation, and weighted regularization triplet loss, we address the inherent challenges of Re-ID, including variations in appearance due to lighting, pose, and occlusion. Our comprehensive approach not only mitigates these issues but also improves feature discrimination and model stability, significantly contributing to the field's progression. Our work offers a robust solution for real-world applications, aligning with the broader goals of public safety and security. The code is available at https://github.com/JDing-JNU/ReID_Human.

External IDs:dblp:conf/icmcs/DingLRCTW24