Loading [a11y]/accessibility-menu.js
Indian Sign Language Translation using Deep Learning | IEEE Conference Publication | IEEE Xplore

Indian Sign Language Translation using Deep Learning


Abstract:

Indian Sign Language is the language used by specially abled population in the Indian subcontinent to communicate with each other. Unfortunately the general population is...Show More

Abstract:

Indian Sign Language is the language used by specially abled population in the Indian subcontinent to communicate with each other. Unfortunately the general population is not aware of the semantics of Indian Sign Language. In this work we present three deep architectures to translate a given video sequence containing the Indian Sign Language sentence to English Language sentence. We have tried to solve this problem using three approaches. First using an LSTM based Sequence to Sequence model(Seq2Seq), second using an LSTM based Seq2Seq model utilising attention, third using an Indian Sign Language Transformer. These models were evaluated on BLEU scores and the transformer model gave a perfect BLEU score of 1.0 on test data.
Date of Conference: 30 September 2021 - 02 October 2021
Date Added to IEEE Xplore: 22 December 2021
ISBN Information:

ISSN Information:

Conference Location: Bangalore, India

Funding Agency:


I. Introduction

Although many advancements have happened in the field of computer vision [1] and natural language processing [2], not much has been done to help the specially abled population to communicate with the general population. Although some work has been done in static gesture recognition and dynamic gesture recognition for Indian Sign Language, nothing has been done for the sentence translation task till date. The work done by Sruthi et al. [3] which is one of the recent works also deals with static gestures. The work done by Neelkamal el al. [4] and Pratik et al. [5] deals with static and dynamic based gestures. This work mainly intends to build a system which can recognise Indian Sign Language sentences from video sequences in real-time and can be trained in an end-to-end manner.

Contact IEEE to Subscribe

References

References is not available for this document.