A Loosely-Coupled GNSS-Visual-Inertial Fusion for State Estimation Based On Optimation

Accurate and high-rate state estimation is the key step for artificial intelligence to be applied to various scenarios. The Visual-Inertial odometer (VIO) can be affected by drift during long periods of operation and cannot obtain real-world positioning trajectoriness. On the other hand, Global Navigation Satellite System(GNSS) can obtain positioning results in the real world, but due to the instability of satellite signals, external interference and other factors, the positioning accuracy is obviously insufficient in many scenes. In this paper, we present a three-sensor fusion system based on nonlinear optimization. The system uses visual, inertial and GNSS information for real-time and drift-free global 6 Degree-of-Freedom (DoF) state estimation. We extensively evaluate the proposed system by real-world experiments, and the result demonstrates that the system makes full use of the advantages of the three types of sensors and can achieve great performance in all kinds of complicated situations including outdoor and indoor environments.