Experimental design for intelligent traffic sign detection in rainy and foggy weather based on YOLOv8

Current traffic engineering laboratory exercises often lack sufficient integration of cutting-edge technologies, have limited experimental design diversity, and result in unequal development of practical skills among students. To address these issues, this experiment was designed for intelligent traffic sign detection within an autonomous driving application. Using the TT100K dataset as a base, the experiment incorporates data augmentation techniques, specifically sign replacement and fog/rain simulation algorithms. Based on the YOLOv8 object detection framework, the experiment guides students through targeted optimizations to better suit traffic sign detection tasks. The effectiveness of the optimized algorithm is validated through ablation studies.

This experiment focuses on building and optimizing a traffic sign detection model using the YOLOv8 architecture. The TT100K dataset was enhanced via data augmentation strategies to improve model robustness in adverse weather. Three key algorithmic improvements were implemented: 1) structural modifications to the C2f module for enhanced feature representation; 2) incorporation of a P2 feature pyramid layer to improve detection performance for small traffic signs; and 3) optimization of the complete intersection over union–based loss function. Ablation experiments were conducted to evaluate the individual and combined contributions of these improvements. Model performance was measured using the mean average precision (mAP) at the intersection over union (IoU) threshold of 0.5 (mAP@0.5) and across IoU thresholds from 0.5 to 0.95 (mAP@0.5:0.95).

Experimental results on the TT100K dataset demonstrated that the mAP@0.5 score increased from 86.40% to 90.90%, and the mAP@0.5:0.95 score improved from 67.60% to 71.00%. These quantitative results confirm the effectiveness of the proposed optimizations. Furthermore, the experimental design successfully cultivated students' practical abilities in image data processing, detection algorithm construction, and model optimization.

This laboratory experiment effectively enhances students' understanding of the working principles and application scenarios of sign detection within intelligent transportation systems. By engaging students in a complete pipeline from data augmentation to model optimization and evaluation, the exercise markedly improved their skills in solving real-world engineering problems. The design successfully integrates contemporary research trends into the curriculum, addresses the limitations of single-method experiments, and promotes the equitable development of practical competencies among students.