A Descriptor with Rotation Invariance of Environment Expression Using Building Boundary of OpenStreetMap

The precise static environment descriptions within high-quality pre-built maps are crucial for the localization of autonomous vehicles utilizing perception-map matching. However, the pre-built map faces significant cost challenges. OpenStreetMap (OSM) provides open access and extensive urban static environment descriptions and has the potential to serve as an alternative approach when the pre-built map is absent. Nevertheless, the environment description provided by OSM presents spatial precision issues that hardly align with the on-board perception. Thus, the perception-OSM association required a method that does not have a high reliance on prior precise spatial information. This paper discusses the environment description method based on building boundary orientations. And the relationship between boundary orientation and the effectiveness of environmental representation with rotation-invariance introduced by boundary orientation and observation direction. To validate the importation of rotation invariance, comparative experiments are conducted on the KITTI datasets. The results indicate that the description method using boundary direction with rotation-invariance can be effectively applied to perception-OSM matching for localization.