Exploring Cross-Domain Techniques in Person Re-Identification: Challenges and Emerging Trends

Cross-domain person re-identification (ReID) has drawn much more attention because of the ability to improve the robustness and versatility of person recognition systems across diverse conditions. This review provides an overview of the modern deep learning methods employed to solve challenges in cross-domain ReID. Considering the level of abstraction of the target domain information, it selects three typical scenarios of infrared images, sketches and text descriptions, and compares classical methods in these three contexts. We not only provide the fundamental paradigms for solving the person re-identification problem, including feature representation learning and metric learning, but also highlight the challenges that exist in different domains and summarize data-efficient and domain-specific solutions. By examining various datasets and evaluation protocols, we identify gaps in current research and propose future directions for improving cross-modality ReID performance. This paper is aimed at providing some resources and pointing out some future trends for researchers looking to advance the field of person re-identification across heterogeneous data sources.