HD-Map Localization

Real-time high-accuracy localization is an extremely challenging task. In autonomous driving and other fields, determining the position of a platform relative to a certain coordinate frame within a given accuracy range is a necessary condition. Here, we establish a localization method based on a geometric-semantic map: deep learning is used to extract environmental elements from a monocular vision sensor and to establish their matching relationship with a pre-built scene-description map, thereby achieving real-time estimation of the platform’s pose.

method diagram