Dynamic Electromyographic Models to Assess Elbow Joint Torque

Abstract

Many studies have investigated the relationship between surface EMG and joint torque. Most studies have used EMG amplitude to assess elbow joint torque with dynamic models. In this paper, we used signal length and normalized zero crossing rates together with EMG amplitude to assess elbow joint torque with EMG-to-Torque models. We compared the performance of single feature EMG-to-Torque models and multi-feature EMG-to-Torque models by calculating the RMS error between estimated torque and true torque. The results show that multi-channel and multiple feature combination is superior to that of the single feature only. In this study, surface EMG signals were recorded from biceps and triceps muscles of 15 subjects. Single-channel and single feature linear model, multi-channel and single-feature model, multi-channel and single-feature model, multi-channel and multi-feature linear model were all used to assess elbow joint torque. The lowest RMS error is 7.6% achieved by four-channel multi-feature 18-order linear model.