Research on Lane Detection Method based on Deep Learning

Inhalt:
The reliability of lane detection directly affects the safety of intelligent vehicles in the process of assisted driving or even unmanned driving. In this paper, three different lane detection algorithms are compared and analyzed, and their detection effects are verified. At present, the mainstream lane detection methods are divided into traditional detection methods based on machine learning and deep learning. Deep learning has more advantages in applicability and reliability than machine learning. In recent years, deep learning has made great progress in vision, and a large number of detection methods have emerged, which has achieved better results than machine learning when applied to lane line detection. In order to clearly display the performance of various detection methods, this paper studies and compares the three semantic segmentation networks U-net, Deeplabv3+ and PSPnet with the help of miou, the mainstream evaluation index. Guide researchers to better choose lane line detection network. Experiments show that Deeplabv3+ network has better detection accuracy than other networks in lane line detection, and also has a good effect in the face of complex scenes.