Person-specific gesture set selection for optimised movement classification from EMG signals

Abstract

Movement classification from electromyography (EMG) signals is a promising vector for improvement of human computer interaction and prosthetic control. Conventional work in this area typically makes use of expert knowledge to select a set of movements a priori and then design classifiers based around these movements. The disadvantage of this approach is that different individuals might have different sets of movements that would lead to high classification accuracy. The novel approach we take here is to instead use a data-driven diagnostic test to select a set of person-specific movements. This new approach leads to an optimised set of movements for a specific person with regards to classification performance.