Key points are not available for this paper at this time.
Improving the performance of autonomous vehicles in complex urban environments is a challenging task and requires the availability of the precise location. Existing localization methods rely heavily on Global Navigation Satellite System (GNSS) implementations with their known gaps, especially in urban canyons. Multipath effects combined with limited access to the sky are a major systematic cause of errors. To overcome this shortcoming, the authors have implemented a multi-localization-source, Kalman-based system, using a combination of GNSS, an Inertial Measurement Unit (IMU), and a novel visual localization method based on landmarks stored in a cloud-hosted local dynamic map. Unique and reappearing signs are automatically detected within a camera frame. The relative position to the vehicle is calculated. By matching these signs with the elements received from the cloud-based local dynamic map, the location of the test vehicle is estimated. Different map-matching strategies are evaluated. The localization can be started with as little as a single sign and the driving direction of the vehicle. Unique landmarks improve the robustness of localization. The position estimation is fused with IMU and GNSS position information to obtain an optimized, reliable position. Therefore, a Kalman filter with multiple inputs was developed. Tests were performed on the test track for automated and connected driving in the City of Hamburg, Germany. The accuracy of the localization was systematically analyzed based on simulations. Performance criteria have been developed to assess both the system and the area under investigation. The measurements performed indicate a significant improvement, especially in situations where the GNSS signals are weak or disturbed. Real live tests proved the functionality of the system, while the simulations suggest an accuracy better than 2 m with four and more landmarks in 95% of the time. The activities are funded within the project "European Digital Dynamic Mapping" (EDDY).
Weltz et al. (Wed,) studied this question.