Monocular Depth Estimation for 3D Map Construction at Underground Parking Structures

Converting the actual scenes into three-dimensional models has inevitably become one of fundamental requirements in autonomous driving. At present, main obstacle to large-scale deployment is high-cost lidar for environment sensing. Monocular depth estimation aims predict scene and construct a 3D map via merely monocular camera. In this paper, we add geometric consistency constraints address non-Lambertian surface problems estimation. We also utilize imaging principles conversion rules produce model from multiple images. built prototype conduct extensive experiments corridor an underground parking structure, results show effectiveness indoor location-based services.