BdSL-Net: A Hybrid CNN-LSTM-Attention Framework for Real Time Bangla Sign Language Recognition

DOI

10.2991/978-94-6239-664-7_90How to use a DOI?

Keywords

Bengali Sign Language (BdSL); Real-time Gesture Recognition; Hybrid Deep Learning Model; CNN-LSTM-Attention Network; Explainable Artificial Intelligence (XAI)

Abstract

Communication barriers between the Deaf and hard-of-hearing (DHH) community and the hearing population in Bangladesh persist due to the lack of automated Bengali Sign Language (BdSL) translation tools. This study proposes BdSL-Net, a real-time BdSL recognition framework based on computer vision and deep learning. A custom dataset of video 2,000 samples covering 40 BdSL signs was developed, from which 1,662 skeletal keypoints were extracted per frame using MediaPipe Holistic. The proposed hybrid neural architecture integrates a 1D Convolutional Neural Network (CNN) for spatial feature extraction, a Long Short-Term Memory (LSTM) for temporal sequence modeling, and an Attention mechanism to highlight the most discriminative motion segments. BdSL-Net achieved 96.08% accuracy and was implemented as a realtime prototype. Explainable AI (XAI) analysis further validated that the model effectively attends to crucial temporal features within gesture sequences. The results demonstrate BdSL-Net’s potential as a visionbased assistive technology for bridging communication gaps and enabling future continuous BdSL translation. The findings confirm that the CNN–LSTM–Attention hybrid model offers high-accuracy recognition of BdSL gestures and provides a viable proof of concept for vision-based assistive communication technologies in low-resource linguistic contexts.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Md Faisal Hasan
AU  - Md. Nazmus Sakib Sheam
AU  - Uzzwal Kumar Biswas
AU  - Jarin Tasnim Tonvi
AU  - Syed Ahsanul Kabir
PY  - 2026
DA  - 2026/06/08
TI  - BdSL-Net: A Hybrid CNN-LSTM-Attention Framework for Real Time Bangla Sign Language Recognition
BT  - Proceedings of the International Conference on Intelligent Data Analysis and Applications (IDAA 2025)
PB  - Atlantis Press
SP  - 1329
EP  - 1344
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-664-7_90
DO  - 10.2991/978-94-6239-664-7_90
ID  - Hasan2026
ER  -