EF2lane: Enhanced Feature Fusion 2D Lane Detection Network In 3d Point Cloud

Lane detection is one of the core tasks in autonomous driving. Lane detection relies primarily on front-view camera images or the bird’s eye view projection of LiDAR point cloud, but it faces challenges in terms of robustness in complex scenarios. In this paper, we propose a 2D lane detection model based on enhanced feature fusion, EF2Lane. EF2Lane extracts lane features from bird’s-eye view of 3D point cloud, and combines them with Mamba like linear attention (MLLA) to capture global correlations. The global features are then fused with the local features extracted by a convolutional network. Additionally, EF2Lane adopts row-wise inference to efficiently process a sparse point cloud. Experimental results demonstrate that EF2Lane achieves superior performance on the K-Lane and CampusLane datasets.