Exclusion of GNSS NLOS Receptions Caused by Dynamic Objects in Heavy Traffic Urban Scenarios Using Real-Time 3D Point Cloud: An Approach without 3D Maps
Weisong Wen, Guohao Zhang and Li-Ta Hsu, The Hong Kong Polytechnic University, Hong Kong
Date/Time: Tuesday, Apr. 24, 11:03 a.m.
Autonomous driving is believed to be a remedy to the excessive traffic accidents. Absolute positioning is a key factor for it. Global Navigation Satellites System (GNSS) receiver is the only on-board device providing absolute localization information. GNSS can supply satisfactory positioning in open or sub-urban areas such as high-way. Unfortunately, its accuracy is greatly challenged in highly urbanized area due to the GNSS signal reflection, diffraction and blockage. The phenomenon is well-known as multipath effects and NLOS receptions that dominating GPS positioning errors. The error can easily reach about 50 meters in deep urban canyon. Therefore, the mitigation of the error caused by multipath and NLOS is the key to facilitate GPS positioning. Designing sophisticated GPS receiver correlator  and antenna-array are general approaches to mitigate the multipath effect. However, these approaches have little improvement on NLOS reception. An eye-catching method to mitigate multipath and NLOS is the integration of GNSS receiver and 3D city models, which is called 3D map aided (3DMA) GNSS . The accuracy of the 3D city model is generally in a meter-level. Thus, the false alarm of multipath detection can arise especially in the occasion that the GPS signal transmits through the edges of surrounding buildings. 3D point cloud-based map is a promising replacement for 3D city models because it is essential for the LiDAR based localization. In addition, the 3D point cloud-based map contains sufficient depth information of surrounding environment represented by 3D point cloud . Integration of GPS, 3D point cloud-base map and LiDAR, using Unscented Kalman Filter (UKF) is implemented among previous research for Unmanned Aerial Vehicle (UAV) application . NLOS is detected and excluded in the GNSS positioning . However, the challenges from multipath and NLOS in the environment of the UAV research are much easier comparing to the environment of road vehicle in mega city such as Hong Kong. Multipath effects and NLOS reception can be caused by dynamic objects on roads in heavy traffic urban scenarios such as the double-decker buses. However, these objects are not modelled in the 3D point cloud-based map or 3D city models. In this case, NLOS reception cannot be detected and excluded by these two kind of 3D maps. The objective of this paper is therefore to improve the GNSS positioning performance for road vehicles in heavy traffic urban scenarios by proposing an approach to detect and exclude the NLOS reception using real-time point cloud. As a result, the improved GNSS positioning result can be estimated using the remaining LOS visible satellites.
The proposed methodology is detailed as the following. Firstly, 3D LiDAR sensor, ranging to 80 meters, is employed to provide real time 3D point cloud representing contour information of surrounding dynamic objects. Then normal distribution transform (NDT) algorithm is utilized to calculate the pose transformation by mapping two consecutive frame 3D point cloud, which is well-known as LiDAR odometry. Real time 3D points are subdivided into cells containing several points. Each cell is modelled by a normal distribution with probability in the algorithm. With the LiDAR odometry, pose prediction of the road vehicle can be obtained continuously. Secondly, a ray-casting algorithm is implemented to simulate the signal transmission routes between GNSS satellites and receiver. The position of satellites is estimated by broadcast ephemeris. The predicted receiver position is provided by the LiDAR odometry. If the line-of-sight (LOS) transmission is blocked by LiDAR sensed 3D point cloud, it means LOS GNSS signal is blocked by dynamic objects or buildings, namely to NLOS reception. Otherwise, the surviving GNSS measurements will be classified to LOS visible measurements. Thirdly, GPS positioning is computed using the surviving pseudoranges by weighted least squares method. Finally, the expected GPS positioning error, which can be used to check the reliability of GPS solution, is calculated by multiplication of the new dilution of precision (DOP) and pseudoranges residues. The new DOP is calculated using the geometric matrix with the identified NLOS measurement. GPS positioning is usually integrated into the GPS/INS/LiDAR sensor fusion framework for vehicle localization, such as the Kalman filtering. Thus, adaptive GPS reliability is essential for further integration of multi-sensors.
The proposed method is verified through real road tests in highly urbanized canyon of Hong Kong. A fiber optic level INS and GNSS integrated navigation system (NovAtel SPANCPT) is used to provide ground truth to evaluate the proposed method. A low-cost GNSS receiver (ublox M8T) is equipped on an automobile to record the GPS and Beidou signals during the driving test. Meanwhile, a 3D LiDAR sensor to gather 3D real time point cloud. According to the experiment results, the GNSS positioning error can be reduced from 50 to 10 meters by the proposed method in the testing environment. In addition, the evaluation of reliability is conducted by comparing the actual and expected positioning error. Our results show that the expected positioning error can capture the trend of actual one in the most of time.
 Fenton, P. (1995). Pseudorandom noise ranging receiver which compensates for multipath distortion by making use of multiple correlator time delay spacing. CA, US 5414729 A.
 Groves, P. D. (2011). Shadow matching: A new GNSS positioning technique for urban canyons. The journal of Navigation, 64(3), 417-430.
 Maier, D., & Kleiner, A. (2010). Improved GPS sensor model for mobile robots in urban terrain. IEEE International Conference on Robotics and Automation (Vol.58, pp.4385-4390). IEEE.
 Shetty, A. P. (2017). GPS-LiDAR sensor fusion aided by 3D city models for UAVs, Doctoral dissertation, University of Illinois.