A multi-stage feature alignment framework with cross-modality collaborative fusion for visible-infrared person...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Visible-Infrared Person Re-identification (VI-ReID) is crucial for intelligent transportation and surveillance.
• Semantic discrepancies between visible and infrared data pose a significant challenge.
• Existing single-stage feature alignment methods struggle with semantic shifts.

Purpose of the Study:

• To propose a Multi-Stage Feature Alignment and Cross-Modality Collaborative Fusion (MS-CF) framework.
• To address the semantic discrepancy challenge in VI-ReID.
• To enhance cross-modality consistency and intra-modality stability.

Main Methods:

• The proposed MS-CF framework utilizes a Dual-Path Cross-Layer Attention (DCA) module for enhanced feature representation.
• A Balanced Feature Normalization (BFN) module improves feature distribution consistency and discriminability.
• A Multi-Stage Hybrid-Modality Alignment (MS-HMA) strategy enables coarse-to-fine semantic convergence.

Main Results:

• The MS-CF framework demonstrated superior performance over state-of-the-art methods on SYSU-MM01 and RegDB datasets.
• Achieved significant improvements of 5.75% in Rank-1 accuracy and 3.97% in mAP on the SYSU-MM01 indoor-search setting.
• Validated the effectiveness of the framework in enhancing robustness and recognition performance.

Conclusions:

• The MS-CF framework effectively mitigates modality discrepancies in VI-ReID.
• Hierarchical feature alignment is key to alleviating semantic shifts.
• The proposed method offers a robust solution for real-world surveillance applications.