Seamless Navigation and Mapping Using INS/GNSS/LiDAR SLAM Multi-Fusion Schemes

Abstract

The sensor fusion techniques have become an integral part of future automated vehicles With advances in sensor technologies it is predictable for the expected growth of sensor fusion in automotive military applications Light Detection and Ranging (LiDAR) is one of the advanced sensor that enables the user to acquire more than 100-meter geospatial measurements with centimeter-level accuracy The concentration of this thesis is to develop the tightly fusion schemes with inertial navigation system/global navigation satellite system (INS/GNSS)/LiDAR and enhance the mapping and positioning performance Conventional mobile mapping systems (MMS) with INS/GNSS and mapping sensors have been widely developed in recent years However current systems and results are still prone to errors especially in GNSS-denied or multipath environments On the other hand simultaneous localization and mapping (SLAM) conducts the positioning and mapping of the environment simultaneously based on the static objects or features surrounded by sensors Precise relative positioning benefits SLAM to continuously localize itself and can further improve the performance when detecting the loop closure on the map However the relative positioning algorithm still suffers from an increase in accumulation error with the traveled distance Moreover once the solution of SLAM diverges it is difficult to re-localize itself on the correct position Therefore this thesis proposes different fusion strategies using INS/GNSS and LiDAR SLAM for unmanned aerial vehicle (UAV) and land vehicle including following novel approaches: (1) hierarchical point cloud registrations combining generalized-iterative closest point (G-ICP) and direct geo-referencing (DG) to address the local minima problem as well as taking the feedback bias into INS/GNSS; (2) kinematic calibration model to estimate mounting parameters used in DG for land vehicle; (3) semi-tightly coupled integration scheme with INS/GNSS/grid-SLAM to achieve the seamless mapping in long-term GNSS-denied environment; (4) navigation algorithms with refreshing map and integrity assessment strategies using INS/GNSS integrating 2D and 3D SLAM The mapping performance of proposed approaches are validated using high accurate reference systems and precise traditional control survey For proposed UAV and land vehicle mapping solutions the generated point cloud and indoor floor plane achieve the meter-level accuracy even in the long-term GNSS outage The positioning performance was assessed based on a large number of trajectories collected by different vehicles and different sensors in various scenarios In general the 3D positioning error can control under 1 5 meter in various scenarios which means it meet the accuracy requirement of “which lane” The results show that the proposed fusion schemes to both aerial and land vehicles have the great potential for future mapping and positioning applications

Date of Award	2020
Original language	English
Supervisor	Kai-Wei Chiang (Supervisor)