Graph-based Map-Aided Localization using Cadastral Maps as Virtual Laser Scans

Environment-based localization algorithms, such as laser odometry, can achieve a remarkable accuracy on the local scale. For autonomous driving, however, it is mandatory to combine these estimates with global information to overcome large scale drift. Our approach uses freely accessible cadastral plans (building fingerprints) together with 2D laser information and odometry in a graph-based approach to realize real-time global localization. The main contributions of our work reside in the way we create a virtual laser scan from cadastral plans, and that we consider the observation integrity by identifying corridor-like environment configurations (ambiguous positioning along the longitudinal axis). Besides, we evaluate our approach on a vehicle in two urban scenarios. We present a comparison of the obtained precision using different relevant combinations of the proposed contributions and show that we can reach an average positioning accuracy of 55 cm at best without requiring a first passage of an equipped vehicle to build a map.