Research Article
Real-Time Sign Language Detection Using Deep Learning Model for Enhanced Communication Accessibility
@INPROCEEDINGS{10.1007/978-3-032-34048-1_4, author={Rita Wiryasaputra and Vivian Angela and Chao-Tung Yang}, title={Real-Time Sign Language Detection Using Deep Learning Model for Enhanced Communication Accessibility}, proceedings={Smart Grid and Internet of Things. 9th EAI International Conference, SGIoT 2025, Taichung, Taiwan, December 6--7, 2025, Proceedings}, proceedings_a={SGIOT}, year={2026}, month={10}, keywords={disabilities healthcare professional machine learning multi classification sign language}, doi={10.1007/978-3-032-34048-1_4} }- Rita Wiryasaputra
Vivian Angela
Chao-Tung Yang
Year: 2026
Real-Time Sign Language Detection Using Deep Learning Model for Enhanced Communication Accessibility
SGIOT
Springer
DOI: 10.1007/978-3-032-34048-1_4
Abstract
The growing prevalence of various diseases has placed people with disabilities at higher risk of health complications. However, despite their increased vulnerability, they continue to experience limited access to proper healthcare service tailored to their needs. Communication remains one of the major barriers, as healthcare professionals often have insufficient knowledge of sign language, making it difficult to provide adequate assistance. Over the years, researchers have attempted to overcome this communication gap by developing and testing models which enable interpreting the alphabet in various sign languages. However, alphabet-level recognition is insufficient to capture the linguistic depth of sign language due to its limited meaning. Therefore, the main purpose of this research is to develop a model that could recognize basic phrases in American sign language (ASL). More than 2,000 data images were collected and classified into 45 classes of basic sign language. A real-time object detection transformer (RF-DETR) model was employed for multi-class classification. The proposed model achieved a precision of 76.6%, demonstrating its potential to recognize basic sign language phrases. Further enhancement may be achieved by employing hybrid models that balance accuracy and latency.