Hand Gesture Recognition Based on Electromyographic and Kinematic Analyses

Abstract

This study investigated the hand gesture recognition methodology based on kinematic and electromyographic analyses. The surface electromyographic (sEMG) signals of 6 muscles were extracted to classify 13 different hand gestures offline to obtain subject's hand movement intention. Dynamic characteristics of the hand were extracted by 25 reflective markers which were attached on the joint of finger. For gesture offline recognition. Totally 16 time domain, frequency domain, time-frequency domain and nonlinear parameters were extracted. We used the support vector machine (SVM) to train classifier and classify hand gestures. There were 18 healthy subjects who participated in this study. Nine of them were selected to analyze the correlation between sEMG signal and thumb, index finger metacarpophalangeal joint. The overall classification accuracy of 13 gestures was 98.45±0.83%. The joint angle of metacarpophalangeal and phalangeal joints was positively correlated with muscle abductor pollicis brevis (APB) at feature median frequency (MDF) and mean power frequency (MPF). The mean absolute value (MAV) and root mean square value (RMS) of the joint angle of thumb metacarpophalangeal joint was positively correlated with muscle flexor carpi radialis (FCR) and muscle APB.