Real-Time Hand Motion Estimation Using EMG Signals with Support Vector Machines

Abstract

Various interfaces using Electromyogram (EMG) signals for controlling a robot hand have been developed. However, there are few researches that apply Support Vector Machines (SVMs) to EMG signal classification for estimating operator's hand motions. There is a possibility that the SVMs are effective classifiers. This paper proposes a real-time hand motion estimation method using the EMG signals with the SVMs. This method consists of two phases for the hand motion estimation. The first phase is the hand motion classification of EMG signal patterns with the SVMs. In addition to amplitude features in the EMG signals, cepstrum coefficients are extracted as frequency features for robust classification. The second phase is the estimation of operator's joint angles. The joint angles are estimated from EMG signals based on simple linear models between the joint angles and the EMG signals. These two phases are designed so that they can be processed in real-time. Experimental results of seven hand motion estimation show the effectiveness of our proposed method.