MFD ‐ YOLO : A Small Traffic Sign Detection Algorithm Applied to the Field of Automated Driving
Weigang Kong, Zenghui Qu, Haiying Liu, Dapeng Liu, Lixia Deng, Fei Lin, Chaoqun Wang, Lida Liu · IEEJ Transactions on Electrical and Electronic Engineering · 2025
As automated driving technology became more widespread globally, traffic sign detection held significant research importance in autonomous driving. Current detection algorithms faced accuracy challenges due to small, deformed, obscured, or weather‐affected targets, resulting in low detection precision. This paper proposed the MFD‐YOLO algorithm, based on YOLOv5s, specifically for traffic sign detection environments. The MFD‐YOLO algorithm designed a novel Multiple Fusion Convolutions module to enhance feature nonlinearity and diversity. It also incorporated a Small Target Detection module within the detection head structure to improve small target detection accuracy while reducing model complexity, thus enhancing deployment capabilities. Additionally, the paper introduced the Bidirectional Feature Pyramid Network architecture, which fused feature maps of different scales to capture multi‐scale information more comprehensively. Extensive validation experiments on the TT100K traffic dataset demonstrated that the MFD‐YOLO algorithm outperformed the original YOLOv5s algorithm, achieving improvements of 9.6% and 8.0% in [email protected] and [email protected]:0.95 metrics, respectively, reaching 72.1% and 54.1%. The results indicated that the MFD‐YOLO algorithm significantly enhanced traffic sign detection, improving safety in autonomous driving scenarios. © 2025 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.