INDIAN SIGN LANGUAGE RECOGNITION USING DEEP LEARNING METHODS

Abstract

This study explores the challenges in using Indian Sign Language (ISL) effectively. We propose using advanced technology to overcome these communication barriers, focusing on improving continuous voice recognition and its translation into ISL. Our approach enhances the accuracy of recognizing continuous sign sequences by integrating a modified Long Short-Term Memory (LSTM) architecture with a Residual Network (ResNet). This combined approach shows superior performance compared to current models, highlighting its effectiveness. We aim to enhance inclusiveness and accessibility for the Deaf population by combining computer vision and machine learning methods. Despite the importance of sign language recognition systems, there has yet to be a comprehensive review and classification scheme. This study fills that gap, providing an academic literature review from 2007 to 2022 and proposing a classification scheme. We reviewed 396 relevant research articles and selected 55 for detailed analysis based on their focus on 15 sign languages and six dimensions: data acquisition techniques, static/dynamic signs, signing mode, single/double-handed signs, classification technique, and recognition rate. Our findings indicate that most research has focused on using cameras on static, isolated, single-handed signs. This study provides a roadmap for future research and knowledge accumulation in sign language recognition. We also reviewed techniques in gesture detection and translation, emphasizing the use of Convolutional Neural Networks (CNNs), Hidden Markov Models (HMMs), and Support Vector Machines (SVMs). Our findings highlight the crucial role of modern technology in enhancing accessibility and inclusion for people with speech impairments, contributing to the development of intelligent sign language recognition systems and fostering a more inclusive society.