A Framework of Ensemble CNN Models for Real-Time Sign Language Translation

American sign language (ASL) is a natural language to minimize communication barrier for the people suffering from hearing and speech impairment. However, sign languages are not a common form of communication in society, and it is lacking a sign language translator to ease the communication. Hence, this research proposes an improved visual-based communication framework to translate ASL alphabets in real-time. The proposed framework consists of different steps i.e., image pre-processing, hand segmentation with U-Net, data-augmentation techniques, and classification of American sign language. A n ensemble classification model i s proposed where V GG19, R esNet-50 and MobileNet are used as base models, and 2 fully connected layers are used as meta model. The proposed approach achieved 99.86% accuracy and performed better when compared with existing literature.