Desarrollo de algoritmo de reconocimiento de movimientos de la mano basado en aprendizaje automático

Abstract

In the field of bionics, projects have been carried out that can perform signal classification, display statistics or control prostheses using machine learning algorithms. The work developed is a contribution to the area of ​​limb rehabilitation, it is an algorithm capable of learning and recognizing the movement of the fingers of the hand individually (little, ring, heart, index and thumb). This article presents the design of the algorithm that processes the signals generated by the arms through an electromyographic sensor (Myo developed by Thalmic Labs). It contains the design of a filtering stage, a training stage and a graphical interface that allows the user to interact with the obtained signals. The project proposes the use of an artificial neural network (multilayer perceptron) using the tools provided by MATLAB software such as the Multi-Layer Perceptron library and the MinGW-w64 Add-On. The results of this development show how the algorithm identifies the differences between the main characteristics of the signals of each finger with an accuracy of 75.71% and uses the collected information to identify new samples. The results of this development show the strengths and weaknesses of the algorithm, and allow to conclude that machine learning algorithms offer a scalable solution for the control of arm prostheses.