Modelo de reconocimiento en tiempo real de gestos de la mano utilizando técnicas de deep learning y señales electromiográficas

Abstract

Gesture recognition has multiple application domains such as medicine, engineering and robotics. It also allows us the development of new and more natural approaches to human-machine interaction. Real-time hand gesture recognition consists of identifying a given gesture performed by the hand at any given moment. In this work, we propose a model for real-time hand gesture recognition, which takes as input the electromyographic (EMG) signals from the muscular activity in the forearm, measured with the Myo Armband sensor. We use an autoencoder for automatic feature extraction, and an artificial neural network (ANN) based on the rectified linear unit (ReLU) transfer function for classification. The proposed model can recognize the same 5 gestures as the proprietary recognize the same 5 gestures as the proprietary recognition system of the Myo Armband (fist, wave left, wave right, fingers spread and double tap), achieving an average recognition accuracy of 85.08%  15.21%, with an average response time of 3  1 ms. The proposed model is general, this implies that it can recognize the gestures from any user, even when their data were not included in the training dataset composed of 50 users. This feature represents an advantage for the proposed model since it does not require training for each user of this model, unlike user-specific models which require training using the data from the person that will use the model.