Key points are not available for this paper at this time.
Pursuing accurate positioning, especially in the lateral direction is a challenging issue in autonomous driving services. However, the commonly used GNSS/INS is often subjected to signal blockages, resulting in rapid divergence for the inertial system in complex urban environments. With the advent and advancement of high-definition (HD) maps, the absolute coordinate information they store can serve as excellent positioning resources. In this case, we propose a real-time and continuous positioning method that fuses multiple information of vehicle-mounted sensors (GNSS, INS, vision, DMI) and HD maps, along with a complete workflow for utilizing commercial HD maps for high-precision positioning. The lane lines are automatically detected from images using a deep learning module, and the corresponding map data is extracted based on the vehicle’s approximate position. Road shape registration is performed to construct measurement equations by iteratively matching the projected and detected lane lines. Finally, the system refines its states with a tightly-coupled INS/DMI/LAN fusion model to acquire reliable positioning results when the GNSS signal is unavailable. Additionally, to address the challenges of encryption offset and missing elevation data in HD maps, an automatic map offset calibration and online height estimation method is proposed using open environment data. The field experiments demonstrate that in a GNSS-denied tunnel environment, the system can maintain centimeter-level accuracy in the lateral direction during a 200-second GNSS signal outage.
Zhu et al. (Mon,) studied this question.