Real-time Detection of Lane Lines and Optimization of Assisted Driving System Based on Improved YOLO Model

The precision, timeliness, and stability of lane line detection have a significant impact on both vehicle safety and the overall driving experience. This study presents a real-time lane detection approach and an optimization strategy for an assisted driving system, leveraging an enhanced You Only Look Once (YOLO) model. Lane markings exhibit elongated and continuous geometric properties. Aiming at this characteristic, the feature extraction network of YOLOv5-L model is optimized. In this study, inverse perspective transformation and data enhancement technology are used to improve the adaptability of the model to complex scenes. Furthermore, modifying the loss function from GIoU to DIoU enhances the accuracy of the predicted bounding box. Experimental results indicate that the optimized YOLO model achieves a mean average precision (mAP) of 93.8% on public datasets, and the frame rate per second (FPS) is 45.6 frames, which is significantly better than classic algorithms such as YOLOv5-L and U-Net. The detection accuracy of the model reaches $\mathbf{9 1. 3 \%, 8 9. 6 \%, 8 6. 8 \%}$ and 83.4% respectively in complex scenes such as illumination change, shadow occlusion, rainy day and night, showing strong robustness. The enhanced YOLO model presented in this study effectively balances accuracy, real-time processing, and robustness, and provides an effective solution for lane line detection and optimization of assisted driving system.