Continuous prediction of shoulder joint angle in real-time

Continuous prediction of dynamic joint angle from surface electromyography (sEMG) signal is one of the most important applications in rehabilitation area for stroke survivors as these can directly reflect the user motor intention. In this study, new shoulder joint angle prediction method in real-time based on the biosignal: sEMG is proposed. Firstly, sEMG to muscle activation model is built up to extract the user intention from contracted muscles and then feed into the extreme learning machine (ELM) to estimate the angle in real-time continuously. The estimated joint angle is then compare with the webcam captured joint angle to analyze the effectiveness of the proposed method. The result reveals that correlation coefficient between actual angle and estimated angle is as high as 0.96 in offline and 0.93 in online mode. In addition, the processing time for the estimation is less than 32ms in both cases which is within the semblance of human natural movements. Therefore, the proposed method is able to predict the user intended movement very well and naturally and hence, it is suitable for real-time applications.