Enhancing BdSL Recognition: Comparative Evaluation of CNN, VGG16, ResNet50, and MobileNet Architectures

DOI

10.2991/978-94-6239-664-7_89How to use a DOI?

Keywords

Bangla Sign Language (BdSL); Deep Learning; CNN; VGG16; ResNet50; MobileNet; Transfer Learning; Gesture Recognition

Abstract

Sign language is an important form of communication for the hearing-impaired. Unfortunately, Bangla Sign Language (BdSL) has been largely neglected in the field of technology studies and applications. We presented a framework to incorporate deep learning into BdSL recognition to create an inclusive means of communication in Bangladesh by translating sign gestures into text in real time. A dataset of more than 3,600 samples was prepared from 36 BdSL gestures in various lighting and hand-shape conditions with the help of 10 volunteers. The data was pre-processed with the following steps: frame extraction, resizing to 224 × 224 pixels, normalization to grayscale, data augmentation, and normalization of landmarks using MediaPipe. Four models were utilized in the study: baseline CNN, VGG16, ResNet50, and MobileNet. The baseline CNN revealed moderate performance, whereas VGG16 with transfer learning bagged higher scores for recognition accuracy. The fine-tuned and augmented MobileNet model outperformed all other techniques at 92% accuracy, which shows its ability to reliably reduce the gesture recognition task. The findings of this study demonstrate the potential for the system to be deployed on mobile devices and embedded systems to create an efficient and effective approach for real-time BdSL-to-text translation for communication with hearing-impaired people in Bangladesh, ultimately supporting their social inclusion.

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  - Jannatul Ferdoss Faria
AU  - Sadia Akter
AU  - Fatema Tuj Tarannom Esty
PY  - 2026
DA  - 2026/06/08
TI  - Enhancing BdSL Recognition: Comparative Evaluation of CNN, VGG16, ResNet50, and MobileNet Architectures
BT  - Proceedings of the International Conference on Intelligent Data Analysis and Applications (IDAA 2025)
PB  - Atlantis Press
SP  - 1317
EP  - 1328
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-664-7_89
DO  - 10.2991/978-94-6239-664-7_89
ID  - Faria2026
ER  -