Long-term re-identification

• Main Author: Eng, Yao Hui
• Other Authors: Alex Chichung Kot
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/167465
• Institution: Nanyang Technological University

Person re-identification (Re-ID) is a problem that has existed for many years and many state of-the-art short-term Re-ID models have been created. However, most of these models focus on short-term Re-ID instead of long-term Re-ID. In recent years, the focus of the community is shifting towards Long-Term Re-ID. For Long-Term Re-ID, the Re-ID models must be insensitive to the subject’s clothing as we assume that changing of clothes are likely to occur. Recently, a long-term Re-ID model, called SPS, with a result almost double its predecessor was shared. Therefore, in this report, I surveyed recent existing long-term clothing-changing Re-ID methods such as the LTCC, PRCC and SPS methods and also the existing dataset available such as LTCC, PRCC and P-DESTRE dataset. I conducted comprehensive evaluations of these methods on these benchmark datasets to analyse their effectiveness and limitations.