An approach of ANN based noise filtering and position estimation for inertial navigation system

Abstract

The inertial navigation system (INS) has been widely used in many areas, however, the readings of the inertia measurement unit (IMU) in the system usually contain noises. In this paper, an artificial multiple-layer neural network is adopted for filtering noises in the acceleration signals from an IMU. The target dataset for the artificial neural network (ANN) is built from the encoder readings of a multiple inputs multiple outputs (MIMO) helicopter system. Three different algorithms namely, Bayesian regularization, Levenberg-Marquardt, and Scaled Conjugate Gradient are used for ANN training. The noise filtering performance is compared to that from the Kalman filter, a common filtering technique. The ANN filtered data and Kalman filtered data are also used to estimate the position of the IMU which is then compared with the position data obtained from the encoders. The comparison results show our approach has a better performance.