Utilizing LiDAR Registration on 3D High Accuracy Digital Maps for Robust Positioning in GNSS Challenging Environments

摘要

The expanding popularity of navigation-based technologies for self-driving cars and autonomous robotics requires an accurate, reliable, and continuous positioning and navigation solution. Global Navigation Satellite System (GNSS) positioning and Dead-Reckoning (DR) navigation techniques are usually combined to provide an integrated solution. However, such an integrated solution can be degraded in GNSS-challenged environments such as urban and downtown environments. In these challenging environments, it is essential to develop new techniques utilizing other available sensor systems to provide the desirable uninterrupted performance. In this paper, we introduce a method that utilizes vehicle motion sensors and LiDAR point cloud registration on high-accuracy 3D maps to provide a navigation solution suitable for GNSS challenging environments. A LiDAR Map Registration (LMR) subsystem is employed to obtain position and attitude fixing information compensating for the drifting errors generated by Vehicle Sensors Dead-Reckoning (VSDR), especially when relying on low-cost sensors. The proposed method is verified using real data from motion sensors and LiDAR point clouds on real road test experiments performed on a land vehicle in downtown Kingston, Ontario, Canada. Additionally, high-accuracy 3D maps of the city of Kingston were employed for the LMR subsystem. Our results indicate the efficacy of the proposed technique, achieving on average an RMSE of 0.86 and 1.29 meters in the horizontal and vertical position components respectively. Our results also showed that we were able to achieve the target of sustainable sub-meter-level and lane-level accuracies 77% and 95% of the time. This indicates the potential of the use of the proposed method as a viable navigation component for modern autonomous vehicles.