Recognition of American Sign Language using Deep Learning

Abstract

Deaf and mute communities have always faced a communication barrier, but advances in the field of Deep Learning are reducing this barrier. As a form of communication, sign language is one of the most ancient and natural, but since few people speak it and interpreters are extremely rare, this paper proposes to use neural networks to handle fingerspelling based on American Sign Language. A comparative study for Sign Language Recognition (SLR) is presented by implementing a variety of Deep Learning models. The paper proposes a CNN architecture that outperforms (by around 4%) various pretrained models for SLR.