Research on a traffic sign detection algorithm based on improved YOLOv12

Abstract

Traffic sign detection is a key component of autonomous-driving perception because missed, delayed, or unstable sign recognition may directly affect vehicle decision-making and driving safety. In practical road scenes, traffic signs are usually small in scale and diverse in category, and they are frequently affected by complex backgrounds, partial occlusion, motion blur, and illumination changes. These factors make conventional detectors prone to missed detections and unstable confidence scores. To address these problems, this paper proposes YOLOv12-lite-Opt, an improved YOLOv12-based traffic sign detector that strengthens C2f feature extraction, optimizes multi-scale feature fusion in the neck, and introduces channel optimization with a lightweight detection head. Experiments on the TT100K dataset show that mAP@0.5 increases from 46.4% to 57.6%, mAP@0.5:0.95 from 34.0% to 42.6%, and Recall from 42.2% to 52.5%, demonstrating improved robustness and practical value for small-target traffic sign detection.