Methodology and comparison of devices for recognition of sign language characters

The purpose of this work is to develop devices capable of identifying sign language characters and comparing them in order to verify layout with better accuracy and robustness. The recognition is performed using Artificial Neural Networks and all the input data are signals from flex sensors, accelerometers and gyroscopes, positioned differently on each device. After being trained, validated and tested, the network reachs hit rate about 95.8%. It is proposed as solution to deaf people's accessibility and presents as layout proposal for the development of new devices for recognition of signals that express complete words and phrases.