Maximum Likelihood Method for Finger Motion Recognition from sEMG Signals

Abstract

We provide a method could recognize finger flexing motions using a 4-channel surface EMG (electromyogram). Surface EMG is harmless to human body and easy to be acquired. But it could not be reflected activity of specific nerves or muscles as compared to the invasive EMG. Accordingly, it was so difficult to discriminate various motions using low number of electrodes. EMG data were obtained from 4 electrodes placed around the forearm. The motions we chosen are flexing of each 5 single fingers (thumb, index finger, middle finger, ring finger, and little fingers) and 3 hand motions like 'O.K. sign'. One object trained these motions enough and another one didn't. The proposed method uses the short-time entropy using windowing function as the criterion for detect activation moment of muscle. Overall information entropy was a measurement that could represent the pattern of muscle activity. Maximum likelihood estimation was used to determine motion models reconstructed with statistical properties of data. Results showed that this method could be useful for recognizing finger motions. The average accuracy was more than 95 %.