Deep Convolutional Real Time Model (DCRTM) for American Sign Language (ASL) Recognition

Abstract

Sign language is a kind of communication rich of expressions, and it has the same properties as spoken languages. In this paper, the authors discuss the use of transfer learning techniques to develop an intelligent system that recognizes American Sign Language. The idea behind was that rather than creating a new model of deep convolutional neural network and spend a lot of time in experimentations, the authors used already pre-trained models to benefit from their advantages. In this study, they used four different models (YOLOv3, real-time model, VGG16, and AlexNet). The obtained results were very encouraging. All of them could recognize more than 90% of images.