A Fast Analytical Two-Stage Initial-Parameters Estimation Method for Monocular-Inertial Navigation

The integrated navigation of the visual and the inertial measurement is becoming a research hotspot in the field of autonomous driving and intelligent navigation. The fusion of heterogeneous sensors can effectively compensate for the deficiency of a single sensor. Therefore, developing a visual-inertial calibration algorithm with good real-time performance, high accuracy, and strong robustness is an urgent issue. The analytical solution-based algorithm can effectively avoid locally optimal solutions during the calibration process and significantly increase the real-time of the system but achieves low accuracy, while the iterative-based calibration algorithm can get high accuracy but sacrifice the running time. In this article, a fast analytical two-stage initial-parameters estimation method for monocular-inertial navigation is proposed. The proposed method introduces the analytical solution method to provide the initial IMU calibration value and avoid the time-consuming problem caused by repeated iteration. In order to solve the problem that the initial estimate value is not accurate, this article adopts the coarse-to-fine strategy, takes the result of the analytical solution as the initial value, constructs the disturbance-related constraints of the parameters, and further improves the precision of the calibration parameters. Furthermore, the proposed method also realizes the online extrinsic transformation calibration, which improves the environmental adaptability of the system. A large number of public datasets experiments, real-world experiments, and comparative experiments prove that the proposed algorithm has a significant improvement in the initialization time and also improves the calibration accuracy to a certain extent, realizing the global sense of real-time online calibration.

Automated summary

The integrated navigation of visual and inertial measurement is a research hotpot, with analytical solution-based algorithms improving real-time performance but sacrificing accuracy, while iterative algorithms achieve high accuracy but at the cost of running time. The proposed method in this study provides a fast initial parameter estimation method for improving both real-time performance and accuracy.