A Traffic Sign Detection Algorithm Based on an Improved YOLOv8n

Traffic sign detection in autonomous driving faces challenges including multi-scale objects, complex backgrounds, and limited edge-computing power. To address insufficient multi-scale feature representation and high false negatives for small traffic signs in YOLOv8n, this study proposes an improved algorithm integrating the VoVGSCSP module with a Multi-scale Contextual Attention (MCA) mechanism. The original C2f module is replaced with VoVGSCSP, enhancing feature representation through parallel residual branches and cross-stage connections. A lightweight neck, SlimNeck, is designed and combined with MCA, employing multi-branch pooling and dynamic weight fusion to capture geometric features and color semantics. The PAN-FPN path is optimized with cross-level connections and learnable weights for adaptive multi-scale fusion. Experiments on the GTSRB dataset show that the improved model reduces parameters to 2.66 M (an 11.6% decrease) and computational complexity to 7.49 GFLOPs, while mAP@0.5 increases from 94.7% to 96.3% and FPS improves from 82.3 to 90.6. The proposed algorithm achieves comprehensive gains in lightweighting, accuracy, and speed, demonstrating its effectiveness and practical applicability.