HD-sEMG Gestures Recognition by SVM Classifier for Controlling Prosthesis

Abstract

Electromyography signals (EMG) are an important source to infer motion intention. It has been broadly applied in human-machine interfacing to control the neurorehabilitation devices such as prosthesis and rehabilitation robot. HD-sEMG is a muscle's activity recorded at the delimited area of the skin using 2D array electrode. This strategy permits the analysis of sEMG signals in both temporal and spatial domain. Recent studies display that the spatial distribution of HD-EMG maps improves the recognition of tasks. This work investigates the use of HD-EMG recording to control upper limb prosthesis. The classification of eight hand gestures of able-bodied subjects was developed. Three feature sets were presented in this work. HOG features, time domain features(TD) and the combination of HOG and average intensity features (AIH). Combination of features possibly improved the performance of the classifier. Results show that the combined of intensity features and HOG features achieved higher performance of classifier than other features (Acc=99.37%, P=98.375%, S=97.5%)