Highly accurate self-localization is necessary to achieve safe autonomous driving. Recently, map-based self-localization has been attracting attention. Maps generally consist of point cloud information and vector information. However, this information is managed separately and is not compatible. Therefore, we propose a map format that bridges point cloud information and vector information. Vector data is extracted for objects in the environment and associated with a normal distribution representing detailed geometric information. The proposed map can be used as a vector map for various tasks in autonomous driving, and the set of normal distributions provides accurate self-localization. Experiments show that the proposed self-localization achieves lane-level accuracy (~25 cm location) in urban Tokyo.
Integrated High-Definition Maps for Self-Localization with Vector and Polygon Data
2022-10-08
674400 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
High-definition maps and localization for road vehicles
Europäisches Patentamt | 2024
|HIGH-DEFINITION MAPS AND LOCALIZATION FOR ROAD VEHICLES
Europäisches Patentamt | 2022
|