Uso da arquitetura YOLO para suportar o reconhecimento de textos braille

Abstract

Braille is a reading and writing system used mainly by people with visual impairments, despite being a system known worldwide, few people who are sighted can read a Braille cell. We believe that the use of computer vision can solve this problem for sighted people, eliminating one of the existing barriers between sighted and blind people. For the development of this work, we carried out a systematic mapping in order to identify which methodologies were being used by the academic community to carry out detection of Braille cells and where these studies are being published, in addition to training neural networks using the YOLO (You Only Look Once) framework with three different databases, DSBI, Donnie Braille and Braille Character Dataset, obtaining an average accuracy of 19%, 100% and 85% respectively.