Reconocimiento de gestos de la mano en tiempo real usando señales electromiográficas y redes neuronales artificiales

Abstract

In this paper, a hand gesture recognition model based on superficial electromyographic signals is proposed. The proposed model is composed of five blocks: data acquisition, preprocessing, feature extraction, classification and postprocessing. For data acquisition, a sliding window approach is applied (using a main window and a sub-window), and the Myo Armband is used to measure the superficial electromyographic signal of the forearm. For preprocessing, a function for rectification, filtering and muscle activity detection is applied to the signal seen through the main window observation. For feature extraction, a feature vector is generated using the preprocessed signal values and the results from the application of a bag of functions. For classification, a feedforward neuronal network is used to classify every segment of the signal seen through a sub-window. Finally, for postprocessing a simple majority voting, based on the labels from the sub-window observations, is applied to assign the corresponding label to the signal seen through the main window. The model responds at an average of 29.38 ms (real time) with a recognition accuracy of 90.7%, which higher than the rate stablished in the state-of-the-art.