In view of an improvement of the real-time and accuracy of lane detection, a detection method based on an improved YOLOv5s model has been proposed under the framework of machine learning. A binarization channel is added to update the data set with the original image after image preprocessing, with anchor-free dded to improve its anchor frame, thus extracting lane line features efficiently. Meanwhile, mixup and mosaic are used to enhance data so as to save GPU memory and enhance the machine's ability to recognize targets, and the loss function is improved to EIOU in order to accelerate the convergence speed and improve the recognition accuracy. Experimental results show that the proposed algorithm can achieve a more accurate lane line detection, and with the mAP increasing by about 30%, its real-time performance and accuracy are much higher than those of YOLOv3 Compared with YOLOv5s its mAP increases by about 11%, thus verifying the good robustness of the improved method.