Pose-Guided Person Re-Identification Under Uniform-Attire Condition
IEEE Access · 2026 · 14 · 168478-168491
DOI: 10.1109/access.2026.3745378 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Yubin Miao und 3 weiteren Autor:innen. Die Arbeit nennt 51 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: National Natural Science Foundation of China, Shanghai Agricultural Science and Technology Innovation Project.
Abstract (Originalsprache)
In scenarios where individuals wear uniform attire—such as operating rooms, schools, or secure facilities—traditional person re-identification (ReID) methods that rely on clothing appearance suffer severe performance degradation. The identical and often loose-fitting uniforms suppress visual distinctiveness and obscure body details. To address these challenges, we construct an integrated cross-modal pipeline that fuses human pose and appearance information. Specifically, we adapt a bottleneck alignment and unidirectional cross-attention mechanism (DSEUA) to filter visual noise under skeletal pose guidance, and incorporate a dual-pooling semantic mining architecture (CSM) to recover fine-grained details. Furthermore, to evaluate under the most extreme condition, we construct a synthetic stress-test dataset via human semantic parsing, where all clothing pixels are replaced with uniform visual noise to simulate identical surgical gowns. Extensive comparisons on three public cloth-changing benchmarks (PRCC, LTCC, and Celebrities-ReID) demonstrate that our pipeline achieves highly competitive performance against existing ReID methods. In addition, controlled ablation experiments on the synthetic uniform-attire stress-test dataset show that pose-guided fusion and dual-pooling semantic mining provide substantial benefits when clothing-induced appearance cues are suppressed. The results indicate that our method successfully mitigates the loss of appearance cues, establishing a robust baseline for identity recognition in severely constrained environments.
Förderung
- National Natural Science Foundation of China
- Shanghai Agricultural Science and Technology Innovation Project
Zitieren
APA
Miao, Y., Ren, X., Li, B., & Luo, J. (2026). Pose-Guided Person Re-Identification Under Uniform-Attire Condition. IEEE Access, 14, 168478-168491. https://doi.org/10.1109/access.2026.3745378
Vancouver
Miao Y, Ren X, Li B, Luo J. Pose-Guided Person Re-Identification Under Uniform-Attire Condition. IEEE Access. 2026;14:168478-168491. doi:10.1109/access.2026.3745378
BibTeX
@article{Miao2026,
author = {Miao, Yubin and Ren, Xuesong and Li, Bolin and Luo, Junjie},
title = {Pose-Guided Person Re-Identification Under Uniform-Attire Condition},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {168478--168491},
doi = {10.1109/access.2026.3745378}
}
EUVIMED