A MEMS IMU Gyroscope Calibration Method Based on Deep Learning
Fengrong Huang, Zhen Wang, Luran Xing, Chunyan Gao
Abstract
The errors of microelectromechanical system (MEMS) inertial measurement units (IMUs) are huge, complex, nonlinear, and time varying. The traditional calibration method based on a linear model for calibration and compensation is obviously not applicable. In this article, a calibration method based on deep learning is proposed for MEMS IMU gyroscopes. In this method, the output model of MEMS IMU gyroscope is constructed by using the temporal convolutional network. Based on the powerful data processing capability of deep learning, the error features are obtained from the gyroscope data in the past, and the gyroscope data after the error compensation can be regressed. The method in this article is validated on public dataset. The experimental results show that compared with the existing MEMS sensor error compensation method based on deep learning, the attitude and position accuracy obtained by the inertial navigation solution using the compensated gyroscope data are improved, which proves that the proposed method can effectively and accurately calibrate the gyroscope error.