Visibility-Guided and Occlusion-Simulated Learning for Robust Person Re-Identification.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Occlusion presents a significant challenge in person re-identification (ReID), degrading feature distinctiveness and matching accuracy.
• Existing methods struggle with partial visibility, necessitating advanced techniques for robust performance.

Purpose of the Study:

• To develop a novel framework, Visibility-Guided and Occlusion-Simulated Learning (VGOSL), for robust person ReID under occlusion.
• To enhance the model's ability to handle partially visible individuals by improving feature representation.

Main Methods:

• Proposed a framework with two key modules: Part-aware Visibility Modeling (PVM) and Occlusion Box Simulation (OBS).
• PVM adaptively reweights local features based on estimated part visibility to guide global representation learning.
• OBS simulates structured occlusions during training using multi-branch supervision to improve robustness.

Main Results:

• The VGOSL framework achieved competitive performance on benchmark datasets including Occluded-DukeMTMC, DukeMTMC-reID, Market-1501, Partial-ReID, and MSMT17.
• Demonstrated effectiveness in both occluded and holistic person re-identification scenarios.
• The approach successfully emphasizes informative regions while downplaying occluded areas.

Conclusions:

• The proposed VGOSL framework offers a robust solution for person re-identification challenges posed by occlusions.
• The combination of visibility modeling and occlusion simulation significantly improves ReID performance.
• The framework's adaptability makes it suitable for diverse real-world ReID applications.