A Novel Attitude Estimation Algorithm Based on EKF-LSTM Fusion Model

Abstract

The application of the low-cost inertial measurement unit (IMU) in many fields is growing, but the related attitude algorithms have the problems of low precision and poor adaptability. In this paper, a novel attitude estimation algorithm based on the fusion model of extend Kalman filter (EKF) and long short-term memory (LSTM) is proposed, which is composed of two main process: the initial attitude estimation of EKF and the subsequent calibration of LSTM. In this algorithm, EKF estimates the target’s attitude angles by the inputs of sensor data from IMU, then LSTM makes a calibration of each axis’ estimated angles, which is weighted with KEF’s result to export the optimal estimation finally. The result of simulation experiment shows that this algorithm is 50.515% lower on average than EKF under different working conditions when using mean squared error (MSE) as the evaluation indicator, which could be concluded that this novel algorithm performs better than EKF, and provides a new way for attitude estimation.