Kineticomyography for Hand Gesture Recognition: Experimental Implementation

Human-machine interfaces (HMIs) play a vital role in medical science and technology by interpreting human voluntary movements, with ElectroMyoGraphy (EMG) being the most widely used signal. This paper explores KineticoMyoGra- phy (KMG) as an innovative alternative, offering greater noise robustness, simpler connectivity, and more intuitive movement control. We evaluated the online implementation of KMG signals for bionic hand control using a 3-axis simulator that mimics muscle movements through three magnets in two different data acquisition scenarios. The experiment investigated five distinct classes of forearm muscle movements, including a rest class, and KMG signals were recorded using 40 sensors across four custom-made PCB boards. After removing signal spikes, various models were trained, and the top five were selected for further evaluation. In online testing, the Support Vector Machine (SVM) achieved 90% accuracy in the first scenario, while the Deep Neural Network (DNN) reached 77% accuracy in the second scenario.