Urban localization based on aerial imagery by correcting projection distortion

Abstract

This study proposes a vehicle localization method that fuses aerial maps and LiDAR measurements in urban canyon environments. The building outlines from an aerial image can be used as appropriate features for matching with the LiDAR data for localization. However, distortions caused by scaled orthographic projection of aerial maps are commonly observed in the images of metropolitan areas, which may significantly degrade the matching and resulting localization performance. In this study, a novel method for correcting such distortions is proposed and used for the vehicle localization by matching the corrected map and LiDAR measurements. Instance and semantic segmentation algorithms were used to distinguish individual buildings and generate corrected outlines of the buildings. A particle filter is applied to determine the pose of the vehicle based on the mutual information between the map and LiDAR measurements. The performance of the proposed algorithm was verified using a dataset obtained in urban areas.