RDCRNet: RGB-T Object Detection Network Based on Cross-Modal Representation Model.

Area of Science:

• Computer Vision
• Machine Learning

Background:

• RGB-thermal object detection uses visible and thermal data for improved robustness, especially in low light.
• Existing methods struggle with precisely registered data and cross-modal distribution differences.

Purpose of the Study:

• To present RDCRNet, a novel framework addressing limitations in RGB-thermal object detection.
• To improve cross-modal feature alignment and information exchange for robust detection.

Main Methods:

• RDCRNet employs a Cross-Modal Representation Model with feature remapping and refinement modules.
• A Cross-Scale Feature Integration Module enhances multiscale detection.
• A self-supervised pretraining strategy leverages masked reconstruction, adversarial learning, and semantic consistency.

Main Results:

• RDCRNet achieves state-of-the-art performance on multiple benchmark datasets.
• The framework demonstrates high computational and storage efficiency.
• Validation confirms superiority and practical effectiveness in real-world scenarios.

Conclusions:

• RDCRNet effectively addresses cross-modal distribution disparities and data scarcity in RGB-thermal object detection.
• The proposed methods significantly improve detection robustness and performance across various object scales.