Lane Line Map Estimation for Visual Alignment

Lane detection is important for visualization-tasks as well as autonomous driving. However, recent approaches have focused principally on the latter part, employing sophisticated sensors. This paper presents a novel lane line map estimation method from single images, which is applicable for visualization tasks such as augmented reality (AR) navigation. Our learning-based approach is designed for sparse lane data under perspective view. It works reliably even in various difficult situations, such as those involving irregular data forms, sensor variations, dynamic environments, and obstacles. We also suggest the visual alignment concept to define visual matching between the estimated lane line map and the corresponding external map, thereby enabling the conversion of various applications related to visualization into score maximization. Experimental results demonstrated that the proposed method could not only be directly used for lane-based 2D data augmentation but also be extended to 3D localization, for viewpoint pose estimation, which is essential for various AR scenarios.