Devices analysis and artificial neural network parameters for sign language recognition

The purpose of this paper is to develop and analyses device capable of identifying sign language. The recognition is performed using Multilayer Perceptron and all the input data are signals from flex sensors, accelerometers and gyroscopes. Artificial Neural Network is tested modifying parameters as: a) number of neurons in only middle layer, b) learning rate between input and middle layers and c) learning rate between middle and output layers. After being trained, validated and tested, the network reachs hit rate about 96.1%. It is proposed as alternative to deaf people's accessibility and solution with good accuracy and low financial cost compared to those devices already on the market.