Sistema de reconhecimento de gestos para aplicação em reconhecimento de língua de sinais.

Abstract

The most commonly performed form of communication by humans is speech. However, it is known that a part of the society is hearing impaired. These people use sign language to communicate with the world around them. The language barrier between spoken and sign language brings marginalization to this minority, hindering their access to leisure, health and education. A static gesture recognition system has been developed in order to identify manual alphabet signs. The system was designed using a pre-trained Convolutional Neural Network as well as a dataset of American Sign Language alphabet images. Experiments were carried out with data augmentation and noise addition for the network training. These experiments’ performance were analysed considering the desired application.