Visual–Inertial–Wheel Tightly Couple GNSS and Barometer With Online Calibration Adopting Differential Model

In automobile positioning, visual–inertial odometry is often fragile and lacks the robustness and precision necessary for accurate positioning. In this article, we propose [GNSS, wheel encoder, and barometer-aided visual inertial navigation system (GWB-VINS)], which adopts a method of nonlinear optimization to tightly couple raw measurements of global navigation satellite system (GNSS), wheel encoder, barometer, and visual and inertial information for real-time and drift-free state estimation. Specifically, we innovatively propose a differential model for GNSS and barometer measurements, accurately modeling the GNSS raw measurements uncertainties by canceling atmospheric effects (such as ionospheric and tropospheric delays) without requiring prior model information. Barometer could provide higher measurement accuracy compared with GNSS by differencing the measurement values between adjacent timestamp in the vertical direction. The wheel encoder helps to overcome some extreme cases, such as pure rotation or fast motion, and is used to enhance the robustness and accuracy of visual–inertial odometry. To reduce error accumulation caused by integration, the preintegration method is used to process raw wheel measurements. We also propose an online calibration method for both the GNSS-IMU extrinsics and wheel–IMU extrinsics. The proposed GWB-VINS is evaluated on extensive large-scale urban driving datasets, with specific verification for each component (i.e., wheel encoder, barometer, and GNSS). GWB-VINS demonstrates competitive performance compared with the existing state-of-the-art methods on publicly available datasets.