A Portable MIDI Controller Using EMG-Based Individual Finger Motion Classification

Classifying the motion of the five fingers of the hand using non-invasive bio-signal readings from the forearm is still an unsolved research challenge. Its solution is relevant to hands-free remote control devices, on-stage live performances, consumer entertainment, the video game industry, and most importantly the design of hand prosthetics for amputees. This paper proposes a solution that uses the continuous wavelet transform (CWT) decompositions of electromyography (EMG) signals from the forearm muscles, and Support Vector Machines (SVM) classification. The resulting design is a low cost, low power and low complexity portable embedded system that is strapped to the arm, where it collects EMG signals, classifies them in real-time, and sends the resulting class labels via Bluetooth to a remote interface. These labels are then converted into musical instrument digital interface (MIDI) commands that can be used to control any MIDI-controllable device. While the design is still at the prototype stage at best, it provides a proof-of-concept of non-invasive finger motion classification solely based on EMG readings from the forearm muscles. Experimental simulation of the expected system achieved 91% accuracy.