An improved YOLOv8 framework fused with attention mechanism and gold-YOLO neck for abnormal driving behavior detection in automobiles

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Vision-based detection of abnormal driving behaviors has become essential in the automated-vehicle era, playing a critical role in safeguarding vehicle occupants. By identifying unsafe actions early, this approach has demonstrated significant improvements in driving safety, accident reduction, and traffic condition management. However, existing research has primarily concentrated on facial expression recognition, overlooking a comprehensive analysis of overall driver postures. Common abnormal driving behaviors targeted in this study include eye closure, yawning, head turning, and phone use. Comparative experiments demonstrate that the Gold-YOLO-v8 variant achieves highly robust and competitive performance. Relative to the baseline YOLO-v8n (precision = 93.496%, F1 = 0.9404, mAP@0.5 = 0.97345), Gold-YOLO-v8 improves precision to 94.387% (+ 0.891%), F1-score to 0.9449 (+ 0.0045), and mAP@0.5 to 0.97598 (+ 0.00253). These results validate that the integration of the Gather-and-Distribute (GD) mechanism provides an optimized architectural balance, effectively capturing fine-grained driver anomaly features while still meeting the real-time detection requirement. After systematic validation of the GD multi-scale fusion architecture, this model is suitable for deployment on lightweight vehicle edge equipment for real-time abnormal driving monitoring, and subsequent work will carry out multi-vehicle real-scene field verification based on richer diversified driving datasets.