Arabic Sign Language Recognition Using Deep Learning Models

Arabic Sign Language (ArSL) is a family of sign languages that are spread throughout the Arab world. This paper focuses on developing a robust deep learning model that is trained on the ArSL2018 dataset to convert images of the ArSL alphabets into Arabic alphabets. The ArSL2018 dataset consists of 54,049 images of 32 alphabets collected from 40 signers. We have implemented and validated several deep learning models, including Convolutional Neural Network (CNN), VGG-16, and ResNet-18. The best result is attained by the modified ResNet-18 model which achieved an average test accuracy of 99.47%.