Sign4all: A Low-Cost Application for Deaf People Communication

Abstract

One of the main barriers for deaf people is communication due to the lack of understanding between them and the hearing society. This fact can considerably affect their daily life by leading to their social exclusion. On the way to an inclusive society, this paper presents a low-cost application to assist people to communicate by means of the Sign Language alphabet. For that, this application includes two functionalities: (1) a Sign Language recognizer, which is in charge of typing the letters being signed by deaf people; and, (2) a virtual avatar signing what is written letter by letter. So, a comparative analysis of different techniques resulted in an accuracy of 79.96% when using the Convolutional Neural Network ResNet50 for sign recognition from an RGB image.