Wavelet-ARMA-Based Short-Term Drift Calibration of IMU Position on Human Lower Limbs

Abstract

Position calibration in the inertial motion capture (Mocap) system is a crux to guarantee the correct calculation of human posture. Due to muscle movement and clothes slippage during motion, the position of inertial measurement units (IMUs) installed on the human body generates displacements. Aiming at the drift error caused by IMUs’ position shift in the short term, this paper proposes a wavelet-ARMA prediction model. Three different wavelet bases, Haar, Daubechies, and Coiflet, are selected to combine with the ARMA model to predict angle sequences, among which the Db4 wavelet-ARMA model with the best prediction performance, choosing the angle sequence predicted by the model and the measurement sequence to set up an error model, and using the Gauss-Newton (GN) and the dynamic weight particle swarm optimization (DWPSO) algorithms to realize IMUs’ position calibration. The experimental results show that the wavelet-ARMA model has better prediction performance than the ARMA, BP, and LSTM models with the value of the calibrated RMSE decreasing. The results confirm that the model proposed in this paper can accurately predict the changing trend of joint angle sequences on human lower limbs during gait motion and automatically calibrate the short-term drift error formed in the process of gait motion.