Attire insensitive person re-identification

• Main Author: Leong, Zaccheus Yu Hao
• Other Authors: Alex Chichung Kot
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/167137
• Institution: Nanyang Technological University

Person re-identification is the matching images of people across disjoint camera views in a multi-camera system. It is useful for public security applications where a watchlist of known people is provided as the target set for searching through a large volume of video surveillance footage where the people on the watchlist are likely to re-appear. In some events (eg., sports events) and locations (eg., central business district), many pedestrians may wear similar clothes (eg. sports uniforms, business suits). Re-identifying a person from a crowd with similar clothing, or the same person at different times in different attire, are challenging problems. In this project, 6 datasets comprising of 3 similar-attire datasets and 2 state-of-the-art methods for attire-insensitive person re-identification were consolidated. Dataset cross-testing capabilities were implemented to evaluate the method's ability to adapt to different training and testing similar-attire datasets. Lastly, the project also carried out 2 experiments for Semantic-guided Pixel Sampling method to improve the network's accuracy.