Research Advanced in Pedestrian Re-identification based on Deep Learning

Abstract

Pedestrian re-identification (re-ID) is always a research hotspot in the intelligent video surveillance community, which has been applied in various fields such as social security and criminal investigation. Though much effort has been devoted to boosting the accuracy and speed, person re-ID still faces numerous challenges and difficulties, including low-quality captured images, inconsistent camera angles, varied pedestrian postures, significant changes in background environments, and issues related to lighting and occlusion. This paper aims to summarize the recent studies of pedestrian re-ID to motivate the subsequent development. Specifically, this paper focuses on the technological development, application needs, and challenges of person re-ID from the aspects of occluded pedestrian re-ID, unsupervised person re-ID, virtual data generation, pose-aware person re-ID, and pedestrian search. It also identifies existing problems and offers a perspective on the future trends. The hope is that through the study of these methods, an understanding of the current development directions in person re-ID can be achieved and researchers can be provided with a reference.