Continuous Finger Joint Angle Estimation With sEMG Signals

Abstract

The current sensing devices for measuring continuous finger movement are either restrictive to users (data glove) or easily influenced by external environment (optical or magnetic trackers based method). Therefore, the objective of this study is developing a continuous finger movement tracking system that is more easy and comfortable to use. The surface electromyography (sEMG) signals applied in this study were collected from human forearm with 10 electrodes, and transmitted to the computer via cables. Timedomain features were extracted and further filtered with a low-pass filter to smooth the features. Three partial least square regression (PLSR) based movement estimation models had been built for the three movements investigated in this study, and one movement recognition model was constructed to determine which movement estimation model would be applied for the new incoming samples. The prediction accuracy evaluated in terms of Pearson’s correlation coefficient ranges from 0.84 to 0.91 for single finger flexion, and ranges from 0.53 to 0.83 for the movement of fingers flexed together in fist. The normalized root mean square error (NRMSE) ranges from 0.04 to 0.1 for single finger flexion, and ranges from 0.046 to 0.14 for the movement of fingers flexed together in fist. The effectiveness of PLSR has also been proved by comparing its performance with linear regression (LR) model.