Adaptive optimization online IMU self-calibration method for visual-inertial navigation systems

•The IMU intrinsics of visual-inertial navigation system need online calibration.•The preintegration of VINS considering the IMU intrinsics are derived.•The optimization weights for IMU intrinsics are adjusted adaptively.•The regularized mahalanobis distance is used for nonlinear optimization.•Experiments show the proposed method outperform the VINS-Mono. The monocular visual-inertial navigation system (VINS) has been widely used for robot navigation, autonomous driving, and augmented reality/virtual reality. However, the low-cost IMU intrinsic parameters consist of the biases, scale factors and misalignment errors are sensitive to the change of environment. To improve the precision of VINS, the IMU intrinsic parameters should be online calibrated, besides sufficient excitation of the motion is required to make them observable. Nevertheless, this condition would not always be satisfied on the whole trajectory. This paper proposed an online IMU self-calibration method, and in order to perform the calibration when there exists sufficient excitation, the adjust parameters are designed to adjust the optimization weights adaptively for IMU intrinsic parameters according to the intensity of the motion. Besides, the regularized Mahalanobis distance is proposed to solve the singular problem for the covariance matrix. Experiments on the public datasets show that the proposed method outperform the VINS-Mono.