A Neural Network Classifier based American Sign Language Recognition System

This paper introduces a communication system for the deaf-mute which helps them translate sign language into computer text using a hand-glove and neural network back-end. The hand-glove is fitted with flex sensors to measure finger movements. The electrical signals generated by the sensors are then processed and transmitted to a computer system with the use of an Arduino microcontroller. The processed signals are fed to a neural network trained to classify 26 letters of the alphabet. Use of neural networks for classification purposes has many significant advantages such as improved accuracy and flexibility of usage in different media. In this project, multiple combinations of feed-forward neural networks were used for the classification task. The network with 23 neurons outperformed rest of the networks with a high classification accuracy, evaluated with a low Mean Square Error of 0.510.