Tightly Coupled Inertial Visual GNSS Solution - Application to LIDAR Mapping in Harsh and Denied GNSS Conditions

Abstract

Global navigation satellite system real-time kinematic (GNSS-RTK) positioning is today a key technology for survey and mapping applications. To extend the capability of GNSS in difficult environment, a tight coupling between GNSS-RTK and an inertial navigation systems can greatly improve the results. This solution is adapted to small GNSS outage under bridges and in urban canyon for automotive survey for instance. If the time spent in GNSS outage is too long or if the kinematic of the survey is too weak, the GNSS inertial solution can be compromised, due to high navigation errors, and ultimately, impossibility to align heading angle at initialization. This occurs most often for pedestrian survey where the GNSS conditions are worst and dynamics are low. Some solutions on the market propose a LIDAR based SLAM to overcome the limitations of INS/GNSS. However, this solution has a great impact on the mapping solution in terms of bill of materials costs, and power consumption. Indeed, two LIDARs are generally used in this case: one dedicated to SLAM and one dedicated to point cloud generation. This paper presents an innovative solution to overcome the GNSS/INS limitations, whereas minimizing the system complexity by using a tightly coupled GNSS/INS solution, coupled with our monocular visual inertial SLAM system (DVM). This solution is capable of initialization in a few seconds, and is very reliable in the long term. This vision/INS/GNSS coupling increases the overall RTK fix rate and broadens the availability of high precision navigation solutions under challenging conditions. In addition, our visual SLAM system can optimize the full visual graph and achieve cm accurate positioning on the full path in indoor, benefiting from GNSS points at the entrance and exit of the indoor survey, as well as visual loop closure. For more flexibility and accuracy, our visual graph optimizer can estimate the intrinsic and extrinsic calibration parameters of the camera using only a few GNSS points, allowing easy third-party camera aiding. Finally, visual inertial SLAM post-processing proposes an alternative to LIDAR SLAM that does not suffer from poor geometry issues. Going further we will assess the performance of our inertial visual GNSS solution by generating a LIDAR point cloud and analyzing the consistency of the point cloud.