Diseño e implementación de un guante inteligente que interpreta lenguaje de señas en tiempo real

Abstract

CONDITIONING OF PROJECT PUBLICATION. Communication between deaf and hearing people is limited by the shortage of interpreters and by the lack of proficiency in ecuadorian sign language. Therefore, this project aims to design and implement a portable system that translates a set of signs into speech using a sensorized glove and a mobile application, serving as a supportive alternative for everyday and educational environments. For its development, a prototype was built with five flex sensors and an inertial unit integrated on an Arduino board; data were collected from multiple users, features were extracted using a multilayer perceptron neural network, and a machine learning model was trained in the Edge Impulse platform. Subsequently, embedded inference and Bluetooth Low Energy communication were integrated with an application that plays audio offline and displays the battery level. In testing, the system achieved high classification accuracy and an average response time of approximately 3 seconds, while also maintaining an affordable implementation cost. In conclusion, the prototype demonstrates the technical feasibility of a low-cost solution for assisted sign-to-speech translation and establishes a scalable foundation for expanding vocabulary and improving recognition robustness. Keywords: inclusion; multiclass classification; neural network; mobile application.