Direct Utilization of LIDAR data in GNSS/IMU Processing for Mobile Mapping Applications
Scanning LIDAR systems have become a standard component in most mobile mapping systems, as they provide an impressive level of detail in 3 dimensions. To obtain real-world coordinates of a LIDAR point cloud, a GNSS/IMU system is often used for the exterior orientation (EO), and post-processing is typically used for the optimal EO estimates. For production of the final point cloud, the navigation EO parameters can either be used directly or in some cases as an initial approximate for subsequent processing with SLAM. The MEMS inertial sensors common in modern systems can have high drift rates and rapid error growth in GNSS-denied environments; therefore, improving the accuracies of the EO values produced by the GNSS/IMU system is highly desirable-even to aid to subsequent processing with SLAM.
This paper looks at the accuracy improvements obtained by adding the scan-by-scan LIDAR matching directly into the GNSS-INS processing workflow. While LIDAR-only matching can be very sensitive to the scene surface geometry and orientation, the coupled LIDAR-inertial-GNSS approach is much more robust. A critical aspect is the sensor-fusion modelling, which this paper addresses. Using both indoor and outdoor datasets collected with the Velodyne VLP16 and HDL32 sensors, the algorithm is shown to significantly improve upon the GNSS/INS only post-processed solution.
It is shown with MEMS-based inertial systems that the combined GNSS/IMU-LIDAR approach can produce sub-metre to sub-50-cm EO results over GNSS outages of up to 30 minutes. Stand-alone GNSS-IMU navigation results in such a scenario may contain many metres of error. Even for typically shorter outages, accuracy improvements are still significant at two or more times with the inclusion of Lidar matching.
Finally, to improve navigation accuracies, zero velocity updates (ZUPTS) are commonly used. Given that LIDAR data provides accurate relative distance, the requirement for regular ZUPTS is mitigated. ZUPTS slow productivity and are difficult to perform with backpack systems. This paper will examine the accuracy cost of eliminating ZUPTS; initial results show that the degradation is quite small, meaning they can be reduced or disregarded with subsequent gains in productivity.