Real-time Knee Joint Angle Estimation Based on Surface Electromyograph and Back Propagation Neural Network

Abstract

Aiming at the human-robot interaction and continuous motion control of lower limb rehabilitation exoskeleton robot, a real-time knee joint angle estimation method based on Surface Electromyograph (sEMG) and Back Propagation Neural Network (BPNN) is proposed. BPNN is used to train and analyze the mapping relationship between sEMG and knee joint angle, upon which a real-time knee joint angle estimation system is built. Moreover, the real-time continuous motion estimation of knee joint angle under normal walking condition is realized. By designing two test environments– offline test and online test, the estimation accuracy, rapidity and robustness of the proposed method are proved. In the offline test, the root mean square error (RMSE) between estimated knee joint angle and true knee joint angle is 5.945° at 1km/h walking speed, 5.486 ° at 2km/h, 4.406° at 3km/h, respectively. In the online test, the RMSE is 6.622° at 1km/h, 5.748° at 2km/h, 5.219° at 3km/h, respectively. By comparing the results with Support vector Regression (SVR) and Long Short-Term Memory Network (LSTM) in both offline and online test, the superiority of the proposed method in continuous motion estimation of knee joint angle is proved.