Análisis de exactitud de reconocimiento gestual aplicando SVM y k-NN en señales EMG

Abstract

The use of the gesture using hands, face, and body positions, establish a form of man-machine communication that has yet to be studied deeply and widely. The purpose of this paper is to illustrate the feasibility of gestural recognition performed with the hand by using a wearable (MYO), which captures the electromyographic (EMG) signals produced by the forearm´s muscles, precisely by forming and maintaining the gesture. The EMG signals are captured using surface electrodes and applied in classifier algorithms to achieve gesture recognition. Twenty-one volunteers participated in this research and two hundred and seventythree gestures were analyzed. Two classifiers were evaluated, namely k-Nearest Neighbor (k-NN) and Support Vector Machines (SVM). SVM-based classifier with Polynomial kernel (97.81%) and Radial kernel (93.03) achieved the best results. A gestural dictionary of hand poses was implemented that can be used for similar research, especially in human-machine control tasks interaction.