Motion discrimination technique by EMG signals using hyper-sphere model

This study developed a method of discriminating real-time motion from electromyogram (EMG) signals. We previously proposed a real-time motion discrimination method using hyper-sphere models that discriminated five motions (open, grasp, pinching, wrist extension, and wrist flexion) above 90% and quickly learned EMG signals. Our method prevents elbow motions from interfering with hand motion discrimination. However, we presume in our method that feature quantities do not change with time. Discrimination accuracy might deteriorate over time. Additionally, our method only discriminated three motions (open, grasp, pinching) for finger motions. This paper proposes the effectiveness of our method for changing feature quantities caused by time variation and a real-time motion discrimination method using new hyper-sphere models for four finger motions (open, grasp, pinching, and 2-5th finger flexion). We carried out two experiments and verified the effectiveness of our method for changing feature quantities and four finger motions discrimination using the new hyper-sphere models.