3D LiDAR aided GNSS positioning and its application in sensor fusion for autonomous vehicles in urban canyons

Abstract

Autonomous driving is well believed to be the potential solution for reducing excessive accidents and alleviating severe traffic congestions. Localization is the key and fundamental part of the robust operation of autonomous vehicles. Moreover, centimeter-level globally referenced positioning is required. Global Navigation Satellite System (GNSS) is currently an indispensable source that can provide absolute positioning information. Satisfactory performance (5~10 meters) can be obtained in open space if a decent sky view is available. However, its performance can be suffered due to the blockage and reflection from environment features in the super-urbanized area, causing the well-known multipath effects and non-line-of-sight (NLOS) receptions. In this dissertation, we present several 3D light detection and ranging (LiDAR) aided GNSS positioning methods which aims to solve the problems caused by GNSS NLOS receptions. Different from the conventional 3D mapping aided GNSS (3DMA GNSS), we proposed to leverage the onboard sensing based on 3D LiDAR to detect the polluted GNSS NLOS signals. Then the detected NLOS signals are excluded, remodeled or even corrected before its application in GNSS positioning or integration with other sensors. First, we present a novel GNSS NLOS exclusion method caused by dynamic objects using LiDAR perception. The surrounding dynamic objects are detected based on 3D point clouds. Then the NLOS signals caused by the blockage from dynamic objects are identified based on the detected dynamic objects. The proposed method relaxes the drawback of the 3DMA GNSS which can not help to detect the NLOS caused by dynamic objects. Second, instead of excluding all the detected GNSS NLOS signals, we introduced a novel NLOS correction method with the aid of 3D LiDAR and building height information. The proposed method moves forward a step and solves the problem of the distortion of satellite geometry distribution caused by excessive NLOS exclusion. Thirdly, a more general solution to make use of the detected GNSS NLOS is proposed by remodeling the detected GNSS NLOS. In addition, the integration of GNSS positioning with NLOS modeling with the LIDAR odometry is presented. Fourth, a sliding window map (SWM) is proposed to remodel or correct the GNSS NLOS signals. The proposed method alleviates the dependence of building height information. In addition, the field of view (FOV) of 3D LiDAR point clouds is significantly enhanced with the help of the SWM. Moreover, a general GNSS signal validation and calibration pipeline are employed to ensure the feasibility of GNSS measurements before its integration with inertial navigation system using state-of-the-art factor graph optimization (FGO). All of the four proposed methods are validated using real data collected in urban canyons of Hong Kong. The improved accuracy shows the effectiveness of the proposed method in GNSS positioning in urban canyons.