Glove-Based Classification of Hand Gestures for Arabic Sign Language Using Faster-CNN

Abstract

Recently, American Sign Language has been widely researched to help disabled people to communicate with others. However; the Arabic Sign Language “ASL” has received much less attention. This paper has proposed a smart glove which has been designed using flex sensors to collect a dataset about hand gestures applying ASL. The dataset is composed of resistance and voltage measurements for the bending of the fingers to represent alpha-numeric characters. The measurements are manipulated using normalization and zero referencing methods to create the dataset. A Convolutional Neural Network ‘CNN’ composed of twenty-one layers is proposed. The dataset is used to train the CNN, and the Accuracy and Loss parameters are used to characterize its success. The dataset is classified with an average success rate of 95% based on the classification accuracy. Loss has decreased from 3 to less than 0.5. The proposed CNN layers have classified ASL characters with a reasonable degree of accuracy.