Lane detection in complex environments: Performance comparison between UNet and VGG16 transfer learning models

Lane detection has gradually become a major part of intelligent transportation with the continuous development of autonomous driving technology. However, in complex and ever-changing environmental conditions, such as rainy or foggy weather, current researches still face many challenges. This study aims to explore lane detection methods in complex environments based on deep learning. The work is carried out using Tusimple as the benchmark dataset to compare and analyze UNet and VGG16 transfer learning methods. The experimental result shows that the UNet model exhibits better performance in various evaluation indicators, with an accuracy of about 98%, precision of about 77%, recall of about 78%, F1 score of about 77%, and IoU of about 63%. Therefore, in research on lane detection in complex environments, the UNet method should be considered as a priority choice. This study offers valuable reference for future researchers in model selection when facing lane detection problem, and looks forward to providing strong support for the development of intelligent transportation systems.