Uniform person re-identification

• Main Author: Li, Dichen
• Other Authors: Alex Chichung Kot
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Person re-identification
• Online Access: https://hdl.handle.net/10356/177687
• Institution: Nanyang Technological University

Person re-identification (person Re-ID) involves matching individuals across different cameras, a process widely applicable in security settings like shopping center surveillance. Technological advancements have led to improvements in person Re-ID, but challenges arise when individuals wear similar attire, impacting matching accuracy. This project focuses on uniform person re-identification, exploring existing datasets and cutting-edge methods. Methods are categorized into human parsing (e.g., Self-Correction for Human Parsing – SCHP) and attention-based approaches (e.g., Head-shoulder Attention Network - HAA). We implement various methods and conduct thorough performance evaluations using benchmark datasets such as White Re-ID, Black Re-ID, and NTUOutdoor-Color. Our assessment considers the strengths and weaknesses of these methods, providing insights for future enhancements.