Robust Person Re-Identification
Solving Re-ID challenges in aerial surveillance, occluded scenes, and cloth-changing scenarios.
Person Re-Identification (Re-ID) is the task of associating images of the same person taken from different cameras. While standard Re-ID has matured, this research focuses on the “In-the-Wild” challenges that prevent deployment in public safety and large-scale urban monitoring.
Research Focus Areas
1. Aerial & Drone-based Re-ID
Unlike fixed CCTV, aerial imagery involves drastic changes in scale, top-down viewpoints, and fast motion. My work focuses on aligning features between ground-level and aerial perspectives to maintain a continuous chain of identity.
2. Cloth-Changing Re-ID
Traditional Re-ID models rely heavily on color and texture (e.g., “blue shirt”). When a subject changes clothes over multiple days, these models fail. I explore shape-based and biometric features that remain invariant to clothing changes, allowing for long-term identity association.
3. Handling Partial Occlusions
In crowded environments (airports, transit hubs), subjects are frequently blocked by pillars, luggage, or other people. We developed mechanisms to dynamically weight visible body parts, ensuring that a partial match is still a reliable match.
References
2025
- Occlusion-aware appearance and shape learning for occluded cloth-changing person re-identificationPattern Analysis and Applications, 2025
2024
- Temporal 3d shape modeling for video-based cloth-changing person re-identificationIn Proceedings of the IEEE/CVF Winter conference on applications of computer vision, 2024
- Unsupervised person re-identification in aerial imageryIn Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024
- Contrastive viewpoint-aware shape learning for long-term person re-identificationIn Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024
- Attention-based shape and gait representations learning for video-based cloth-changing person re-identificationarXiv preprint arXiv:2402.03716, 2024
- Contrastive clothing and pose generation for cloth-changing person re-identificationIn Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024
- Tackling domain shifts in person re-identification: A survey and analysisIn Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024
- Occluded cloth-changing person re-identification via occlusion-aware appearance and shape reasoningIn 2024 IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS), 2024
- Acml: Attention-based cross-modality learning for cloth-changing and occluded person re-identificationIn 2024 IEEE International Conference on Image Processing (ICIP), 2024
- Occlusion-aware cross-attention fusion for video-based occluded cloth-changing person re-identificationIn 2024 IEEE international joint conference on biometrics (IJCB), 2024
- Crossvit-reid: Cross-attention vision transformer for occluded cloth-changing person re-identificationIn Proceedings of the Asian Conference on Computer Vision, 2024
- Cross-modality complementary learning for video-based cloth-changing person re-identificationIn Proceedings of the Asian Conference on Computer Vision, 2024
2022
- Unsupervised person re-identification based on skeleton joints using graph convolutional networksIn International Conference on Image Analysis and Processing, 2022