Real-time motion discrimination considering variation of EMG signals associated with lapse of time.

This study proposes a motion discrimination method that considers the variation of electromyogram (EMG) signals associated with a lapse of time. In a previous study, we proposed a real-time discrimination method based on EMG signals of the forearm. Our method uses a hypersphere model as a discriminator. In motion discrimination using EMG signals, one problem is to maintain high discrimination accuracy over time because EMG signals change with a lapse of time. This study analyzed the effect of changes in EMG signals on our method. Based on analysis results, adding a relearning system of the decision criteria to the discrimination system was expected to be effective. We created a new motion discrimination method that contains the relearning system and experimentally verified its effectiveness. The motion discrimination system discriminated three hand motions, open, grasp, and pinch with discrimination accuracy above 90% in real-time (processing time below 300 ms) even after time elapsed.