EALaneNet: Enhancing Lane Detection with Edge-Aware Unit and Curvature-Optimized Loss

Lane detection plays a crucial role in intelligent driving and smart transportation systems. However, existing methods often overlook edge information in images, leading to insufficient detection accuracy in complex scenarios. To address this issue, we propose a novel lane detection method that incorporates an Edge Aware Unit (EAU). By integrating the EAU, we effectively utilize edge information in images and embed prior knowledge into the model, thereby enhancing lane detection accuracy. Additionally, to tackle the challenge of low regression accuracy for points in curve sections, we design a weighted smooth L1 loss function. This method applies an exponential decay function to weight the smooth L1 loss, assigning weights based on the difficulty of the regression points. By increasing the weights for points in complex curve sections, we improve the model's adaptability to challenging curvatures. We conducted experiments on the CULane dataset and our private lane detecion dataset LM-Lane, and the results demonstrate that the proposed method not only excels in standard lane detection tasks but also shows significant advantages in handling curve scene.