I. Introduction
Over the past few years, autonomous driving has received lots of interest from both academic and industry. As a fundamental function of the advanced driver assistance system (ADAS) [1], [2], lane detection has always been a hot topic in the research of autonomous driving technology. It is essential in applications such as vehicle real-time positioning [3], lane keeping assist [4], and forward collision warning [5]. As deep learning has made substantial progress in computer vision research, the corresponding methods have also been applied to lane detection, and compared with traditional algorithms, it has more robustness and higher accuracy. However, there are occlusions, light fluctuations, and weather changes in the real-world traffic environment, which still make lane detection a very challenging task.