Person Re-identification Method Based on Video Spatial Feature Enhancement

• Published in: 2024 IEEE International Conference on Cognitive Computing and Complex Data (ICCD) pp. 23 - 30
• Main Authors: Ke, Zunwang, Sun, Guozhi, Guo, Run, Du, Minghua, Zhang, Yugui
• Format: Conference Proceeding
• Language: English
• Published: IEEE 28.09.2024
• Subjects: attention; Data mining; feature enhancement; Feature extraction; Identification of persons; Monitoring; Pedestrian occlusion; Pedestrians; Person re-identification; Public security; Security; Source coding; spatial features; Splicing; Time-domain analysis
• DOI: 10.1109/ICCD62811.2024.10843613

Recently, video-based person re-identification has received more and more attention, and has played a very important role in public safety fields such as security monitoring and public security criminal investigation. However, it is still a great challenge for existing work to effectively overcome the occlusion problem and extract more abundant pedestrian spatial feature information. To tackle the problem of occlusion of pedestrians in real scenes, this paper proposes a person re-identification method based on video spatial feature enhancement. This method enhances the spatial information in video frames by using multi-angle feature aggregation of time attention and extracts continuous pedestrian detail features by using multi-frame spatial feature splicing of time domain information, so as to solve the difficult problem of pedestrian occlusion recognition. Experiments show that the proposed method has improved performance on MARS and DukeMTMC-VideoReID datasets, which verifies the effectiveness of the proposed method. The source code is available at https://github.com/sgzhi11/Video-Spatial-Feature-Enhancement.