Enhancing navigation performance through visual-inertial odometry in GNSS-degraded environment

Abstract

In recent years, with the rapid development of automated driving technology, the task for achieving continuous, dependable, and high-precision vehicle navigation becomes crucial. The integration of the global navigation satellite system (GNSS) and inertial navigation system (INS), as a proven technology, is confined by the grade of inertial measurement unit and time-increasing INS errors during GNSS outages. Meanwhile, the ability of simultaneous localization and environment perception makes the vision-based navigation technology yield excellent results. Nevertheless, such methods still have to rely on global navigation results to eliminate the accumulation of errors because of the limitation of loop closing. In this case, we proposed a GNSS/INS/Vision integrated solution to provide robust and continuous navigation output in complex driving conditions, especially for the GNSS-degraded environment. Raw observations of multi-GNSS are used to construct double-differenced equations for global navigation estimation, and a tightly coupled extended Kalman filter-based visual-inertial method is applied to achieve high-accuracy local pose. The integrated system was evaluated in experimental validation by both the GNSS outage simulation and vehicular field experiments in different GNSS availability situations. The results indicate that the GNSS navigation performance is significantly improved comparing to the GNSS/INS loosely coupled solution in the GNSS-challenged environment.