On the development of Robust Longitudinal Person Re-Identification

点击翻译按钮获取中文摘要

As one of the most rapidly emerging research areas, person re-identification (Re-ID) has been increasingly receiving attention in recent years. Based on the employed modality in question, the research on Re-ID could be broadly categorised into three branches: 1) retrieving a person of interest across multiple non-overlapping cameras, 2) re-identifying an individual based on a series of video clips, 3) establishing the longitudinally robust templates for biosignal-based person recognition.This project is devoted to thoroughly exploring the person Re-ID and further push the line within the selected research direction. With the advancement of deep neural networks and the increasing demand of intelligent video surveillance, the proposed question has gained significantly increased interest in the computer vision community in particular. Thanks to the availability of publicly databases, the successful candidate would potentially be trained to master both image and time-series data wrangling, which paves the way for the student with a much broader career perspective. One of the import papers related to this project is listed below [1]:[1] M. Ye, J. Shen, G. Lin, T. Xiang, L. Shao, and S. C. H. Hoi, "Deep Learning for Person Re-identification: A Survey and Outlook". IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021.