3D LiDAR Aided GNSS Real-time Kinematic Positioning via Coarse-to-fine Batch Optimization for High Accuracy Mapping in Dense Urban Canyons

Global navigation satellite system (GNSS) real-time kinematic (RTK) has shown centimeter-level absolute positioning results in open-sky areas. However, it is also known to suffer from polluted GNSS measurements and poor satellite geometry in urban canyons, because of the non-line-of-sight and multipath reception caused by the signal blockage and reflection. Light detection and ranging (LiDAR) sensors, along with LiDAR/inertial measurement unit (IMU) odometry systems, can provide precise environment description and short-term accurate relative positioning capability, which could be utilized for aiding GNSS-RTK to obtain better performance. The recently developed 3D LiDAR-aided GNSS-RTK positioning methods detected the GNSS NLOS receptions via the incrementally built map and improve the satellite geometry using the low-lying virtual satellite from LiDAR features. However, the high-elevation angle NLOS receptions can not be fully detected and the multipath signals cannot be effectively mitigated. To fill this gap, this paper proposes a 3D LiDAR aided GNSS-RTK positioning method via iterated coarse to fine batch optimization by (1) Global 3D point cloud map aided NLOS exclusion which enables the detection of high-elevation angle NLOS receptions. (2) Iterated batch optimization based on a innovatively devised tightly-coupled factor graph which fully exploits the global consistency among the constraints of LiDAR, IMU and GNSS-RTK to exclude the potential multipath signals. Our proposed method aims to achieve lifelong accurate positioning performance in deeply urbanized areas. The effectiveness of the proposed method has been proved by the evaluation conducted on our open-source challenging dataset, UrbanNav, which contains various sequences collected by automobile-level low-cost GNSS receivers in urban canyons of Hong Kong.