Classification of forearm and finger motions using electromyogram and arm-shape-changes

Robot arms for humanoid are widely developed for medical, welfare and education use. Surface electromyogram (sEMG) signals which are the electrical signals obtained from surface of human skin using electrodes have been mainly used for classification of hand motions. However, it is difficult to classify detailed motions such as finger motions and wrist pronation or supination. Moreover, Kinect is an integration sensor device which can capture human joints movement. It also has been widely used for recognition of body motions in many fields. However, it has some problems such as setting of camera and restriction of detection range. In this study, we propose an advanced method of motion classification by combining arm-shape-changes with sEMG to classify the detailed motions. Arm-shape-changes are forearm deformation caused by a bulge of muscle when subjects move an arm or a finger. Experimental results showed classification accuracies of 90% or more in wrist pronation and supination which are difficult to classify using only sEMG signals. As the result, our method could classify the detailed motions and contribute to expansion of classifiable hand motions.