Implementación de un algoritmo para el reconocimiento de gestos de la mano en tiempo real, usando señales electromiográficas

Abstract

In this document we propose a recognition system of 5 hand gestures (fist, wave in, wave out, open and pinch), based on 8 channels of Electromyography (EMG), using the Myo armband device placed on the forearm. Traditionally, the preprocessing and feature extraction stages are performed without a methodology that justifies them or shows their influence on the accuracy and precision of classification and the highest importance is assigned to the classification method. In our case, in general, if more than three variables are selected, their experimental values will form clouds or clusters of points in n-dimensional space. Thus, now the selection of those algorithms is based on maximizing the Euclidean distance between clusters and minimizing the dispersion of these clouds. On the other hand, a much neglected matter is the understanding of the phenomenon based on both intrapersonal and interpersonal variability, considering the sex, training degree of the subject and physical makeup. Finally, after understanding these scenarios, a methodology is proposed which, applied to a group of 12 subjects, obtained a classification accuracy of the 5 gestures of 97%