3DLMF: Monocular 3D Lane Detection Based on MetaFormer

In autonomous driving, 3D lane detection using a monocular camera has always been one of the significant challenges, especially in complex scenarios such as uphill and downhill situations. To address these issues, we propose the 3DLMF algorithm, which utilizes the MobileMF network for feature extraction to achieve 2D lane detection. Reconstructing the lane enhances spatial information, while the new loss function optimizes detection performance. The results show that 3DLMF improves the F1 score on the Openlane dataset by 13.0% for 2D lane detection and 2.4% for 3D, demonstrating strong performance.