Automatic Generation of Road Feature Identification Models from Point Cloud Data Using HD Maps

A large amount of point cloud data has been measured and accumulated at various locations for Japanese public works by introducing mobile mapping systems and terrestrial laser scanners. However, since point cloud data are a huge set of points that hold position coordinates, their use is limited in the unprocessed state. For this reason, we have been proposing a method for constructing a product model that structures point cloud data by assigning the meaning of a planimetric feature to the points using a plan of completion drawing that shows the completed shape of a construction object and 3D map data with high precision for autonomous driving. However, the applicable scope of the existing method is limited to the plan of completion drawing and the already developed sections of road maps. Therefore, a method of extracting road features from point cloud data using deep learning has been proposed. However, it requires manual preparation of a huge amount and highly qualified training data. In this study, we propose a method to automatically generate training data for constructing a road feature identification model from point cloud data automatically extracted using road maps. The usefulness of the proposed method has been confirmed through demonstration experiments.