Classificação de libras em imagens através de redes neurais convolucionais

Abstract

The literacy of deaf people is currently a major challenge that has generated many discussions in the educational context. Despite the existence of sign language (Libras), the training of people to carry out teaching activities in this area is quite scarce, and this has made the teaching and learning process of people with or without hearing problems very difficult. With the increasing growth of the internet and computer technologies, it was needed to create advanced artificial intelligence applications to improve this process. A very important milestone was the emergence of Computer Vision, which is an area of artificial intelligence that seeks to analyze, interpret and extract certain useful information from images, also studying the use of emotions, recognition, and analysis of interactive movements of human beings through computerized AI systems that simulate the thoughts and actions of human beings. The main subject of this work addresses a research related to the application of convolutional (or deep) neural network, which are similarly linked to computer vision. Experiments were carried out using a Libras database that served as support for the training of images through the YOLOv5 deep neural network algorithm to perform the Classification of some groups of 6 and 7 images. Subsequently, the results of the tests were compared, and it was observed which periods had a better performance during the classification of Libras images