Cloth-Changing Person Re-identification: A Survey on Challenges, Methods, and Frontier Advances

Person re-identification is a cornerstone of intelligent surveillance systems. However, conventional approaches heavily rely on clothing cues, which leads to severe performance degradation in long-term monitoring scenarios where individuals change outfits. This paper presents a comprehensive survey of cloth-changing person re-identification (CC-ReID), a highly challenging and emerging research frontier, with the goal of systematically summarizing how to distill stable identity essence from appearance attributes that are intrinsically mutable. We first propose a hierarchical taxonomy of CC-ReID methodologies, covering RGB-based implicit feature disentanglement, multimodal assistance leveraging skeleton and gait cues, fine-grained modeling based on human parsing, and recent paradigms that integrate generative models with vision-language pretraining. We then curate representative datasets and evaluation protocols, and conduct cross-method comparisons on benchmarks such as PRCC and LTCC to reveal the performance divergence and generalization bottlenecks of different technical routes under real-world complexity. Finally, we discuss outstanding challenges, including computational overhead and data scarcity, and highlight future directions in multimodal large-model collaboration as well as open-world unsupervised and lifelong learning for achieving practical breakthroughs.