GVI-Switch: A High-Precision GNSS-Visual-Inertial State Estimator for Indoor-Outdoor Environments

Abstract

We propose a high-precision GNSS-visual-inertial state estimator named GVI-Switch, augmented with a loosely coupled Error State Kalman Filter (ESKF), to achieve stable six-degree-of-freedom (6-DoF) pose estimation under varying GNSS signal conditions. First, we developed a GNSS degradation detection scheme based on commonly observed GNSS degradation phenomena. This scheme effectively identifies and eliminates abnormal GNSS signals. During the joint initialization phase, we account for the noise amplification in GNSS signals caused by the carrier's passive movement and propose a method to detect and exclude signals that are severely affected by noise. In the loosely coupled ESKF process, to handle GNSS signal fluctuations during transitions between indoor and outdoor environments, we designed a feedback buffering mechanism that constrains the error state vector. This mechanism not only mitigates the accumulation of errors in a degenerate system but also ensures the stability of positioning results during GNSS status changes. Experimental results demonstrate that our algorithm achieves high-precision, stable, and continuous global positioning outdoors, with an Absolute Trajectory Error (ATE) reduction of up to 70% compared to state-of-the-art (SoTA) GNSS fusion algorithms.