Fast monocular visual-inertial calibration and initialization with weak environmental control and motion control

[weisongwen]

Most existing visual-inertial systems require control of extrinsic environment and extrinsic motion to obtain calibration results and simplify the initialization model. The former limits the usage scenarios of the method, while the latter reduces the estimation accuracy of the state parameters of the navigation system. Therefore, a visual-inertial calibration and initialization method is necessary for weak environmental and motion control conditions. Firstly, to overcome the reliance on environmental control, we establish a motion constraint equation based on the relative consistency of visual-inertial sensing spatial motion and implement a visual-inertial online calibration method. Secondly, to eliminate the reliance on motion control, a dual graph optimization structure is be constructed. The initial optimization outcome serves as the foundation for achieving improved calibration results in the refined optimization. Thirdly, to improve initialization accuracy, a complete initial state model is considered to decouple and calculate the translation extrinsic parameters. These parameters are compensated into subsequent initialization, enhancing the robustness of the initialization process. The proposed method is validated on the EuRoC dataset, and experimental results demonstrate that it achieves more accurate calibration results at a faster rate than VINS-Mono in scenarios without calibration target and with limited vehicle motion. The navigation results of different scenarios demonstrate that our method can more effectively integrate visual-inertial data due to improved calibration and initialization estimates.