Motion estimation of elbow joint from sEMG using continuous wavelet transform and back propagation neural networks

Estimation of continuous joint movement of human arm is significant in human-robot interaction. In this study, we proposed an estimating method combined continuous wavelet transform (CWT) and back propagation neural networks (BPNN) to estimate the joint angles and velocities of human arm from surface electromyography (sEMG) signals during the human-robot interaction. The proposed method was evaluated by comparing to other existing methods and the results derived by using a goniometer through experiments. Nine conditions with different motion patterns and force levels are conducted in the experiments. The root mean squared error (RMSE) and the correlation coefficient (CC) were used as the evaluation indices. Experimental results showed that the estimated angles and velocities of elbow joint were well match to the "true" values derived by the goniometer. The average RMSE of estimated angle and velocity derived by the proposed method was 8.78 degree and 9.59 degree/second, respectively. The average RMSE in both the joint angle and velocity for the proposed method was the smallest comparing to the back propagation neural networks, the higher-order polynomial, and the support vector regression.