Comparative Study and Performance Analysis of Deep Neural Networks for Sign Language Recognition using Transfer Learning

Abstract

Listen

Translate article

Sign Language is the primary source of communication for people with hearing disabilities. Communicating with people with such a disability is a challenge. An Automatic Sign Language Recognition system will reduce this communication gap and facilitate ease of conversing with people with hearing disabilities. In this work, we have compared the performance of 6 popular and recent deep neural network architectures for sign language recognition, namely DarkNet19, DarkNet53, InceptionResNetv2, ResNet50, DenseNet201 and VGG-19. Among these, DarkNet53 performed the best with our self-created dataset of 16,000 images from 60 classes with an average accuracy of 98.02%.