Year: 2025
Type: article
Abstract: Pedestrian detection is crucial for autonomous driving safety, yet single-modal approaches (e.g., RGB) struggle in low light and adverse weather. To address this, we design a Cross-Modality Fusion Transformer (CFT)-based pedestrian detection system integrating visible (RGB) and infrared (TIR) images for accurate recognition and real-time alerts. The CFT method leverages Transformer self-attention for intra- and inter-modal feature fusion, overcoming CNNs' limitations in long-range dependency modeling and cross-modal correlation. The software, built with PyQt5 and OpenCV, supports video preprocessing, YOLOv5-CFT-based detection, and real-time processing via multi-threading. Experimental results confirm significant detection accuracy and robustness improvements in challenging conditions. Additional features include user management, record saving, and screenshot capture, enhancing usability. This research validates CFT's feasibility in autonomous pedestrian detection, providing a foundation for multi-sensor fusion advancements.
Author: Sixu Fei
Institution: Southeast University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4409237158