A Comprehensive Review on Group Re-identification in Surveillance Videos

Abstract

Computer vision plays an important role in the automated analysis of human groups. The appearance of human groups has been studied for various reasons, including detection, identification, tracking, and re-identification. Person re-identification has been studied extensively over the last decade. Despite significant efforts by the computer vision research community, person re-identification often suffers from issues such as similar clothing appearances, occlusion, viewpoint changes, etc. On the contrary, group re-identification has not received much attention. It involves identifying human groups across multiple non-overlapping camera views. It is a challenging problem that suffers from issues related to person re-identification and additional challenges like variations in the number of persons, the structural layout of groups, etc. This paper summarises the research paradigms of human group analysis. It reviews the recent advancements in group re-identification, including key challenges, datasets, and state-of-the-art methods. The paper concludes with a discussion of open research challenges and future directions in group re-identification, including the need for reliable techniques, varied datasets, and ethical considerations regarding privacy. Overall, this paper offers a thorough and up-to-date summary of the most recent findings in group re-identification. It also identifies the research gaps as placeholders for further study.