Bangla Sign Language Detection in Real-Time by Using Action Recognition and LSTM Deep Learning Model

Effective communication is vital for promoting inclusivity and understanding in all areas of life, especially for individuals with hard of hearing, deaf or mute. One of the main ways that these people communicate visually is through sign language. In this study a novel approach of sign language recognition has been executed on a newly developed Bangla Sign Language (BSL) dataset compromising 27 actions totaling 24,300 images. To recognize the sequence of movements and classify them into the relevant BSL actions, these characteristics are fed into an LSTM network. Tested on a large and effective dataset of BSL, the proposed method outperforms conventional techniques with a 95.01 % accuracy rate. These results demonstrate the possibility of deep learning for precise and very effective BSL recognition, which will help people who speak Bangla but have speech and hearing impairments communicate more effectively.