Real-Time Surface EMG Pattern Recognition for Shoulder Motions Based on Support Vector Machine

Surface electromyography (sEMG) signals contain humans' motion intentions and can be used for intuitive control of prostheses or exoskeleton. Although recent research proposes several pattern recognition methods based on sEMG and reported high accuracy, the real-time applications are still limited due to the relatively low accuracy and long time consumption. In this paper, we propose a real-time shoulder motion pattern recognition model based on surface electromyography (sEMG). The Delsys Trigno wireless EMG system with customized LabVIEW program is applied to acquire surface EMG generated by shoulder-related muscles during different shoulder motions. Surface EMG features were extracted and used to create motion recognition. Support Vector Machine (SVM) model was selected and trained for real-time motion recognition. Motion recognition results were given every 135 milliseconds. In order to evaluate the model, an experiment with four shoulder related motions was set up and conducted on five subjects. The experiment result shows that the average accuracy can reach 87.6% in offline training and 85.3% in real-time validation.