Person re-identification for similar clothing or uniform problem

• Main Author: Meng, Zexin
• Other Authors: Alex Chichung Kot
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/163987
• Institution: Nanyang Technological University

Person re-identification (Re-ID) is a technique to retrieve the specific pedestrian in an image or video sequence based on general characteristics of the human body. The information on pedestrian clothing has a significant impact on person Re-ID results. However, in some particular cases (e.g., hospitals, schools, and teams, where everyone must dress uniformly), the clothing features of the target person in the query dataset and other pedestrians in the gallery are almost identical. This brings many challenges to the current person Re-ID methods challenging under similar clothing scenarios. To improve the accuracy in this scenario, this dissertation contrasts the semantic segmentation model with the head-shoulder adaptive attention network (HAA), and the most robust semantic segmentation model was found for the uniform class dataset. Finally, the model has extensively experimented on the datasets Black-reID and NTUOutdoors, with Rank1 significantly outperforming earlier techniques.