Work

Advancing disability-inclusive communication with DynaFuseNet and transformer models for sign language interpretation

Year: 2026

Type: article

Abstract: Fostering inclusion and overcoming accessibility barriers necessitates effective communication for individuals with disabilities. This study demonstrates the integration of Transformer models with a proposed deep learning architecture, DynaFuseNet, to interpret sign language. DynaFuseNet is a composite model that is capable of processing both static and dynamic sign representations. It is constructed by combining a FAST-SIFT Feature Fusion Framework (FSF 3 ) with a Dynamic Temporal Convolution Network (DTCN). The Transformer will enhance the system’s ability to capture long-range dependencies in temporal data, while the architecture employs multi-task learning to analyze various sign language structures simultaneously. To further enhance accessibility, the model will integrate natural language processing (NLP) techniques that would seamlessly translate sign languages. To ensure adaptability and inclusivity, rehabilitation centers have collaborated to create a comprehensive dataset that encompasses a range of regional sign languages and signing techniques. The WLASL dataset achieves 99.12% accuracy, while the MS-ASL dataset achieves 98.56% accuracy. The proposed system obtains state-of-the-art performance. The benefits of the DynaFuseNet architecture and its integration with Transformers in the pursuit of new frontiers in sign language recognition are evident in these results. Thus, this research substantiates the transformative potential of artificial intelligence (AI)-driven assistive technologies in fostering inclusive communication and empowering individuals with disabilities, thereby bridging the linguistic and accessibility disparities within and across diverse communities.

Source: PeerJ Computer Science

DOI: 10.7717/peerj-cs.3543

Authors: Hooria Shahbaz, Armughan Ali, Abdullah Rashid, Salah Uddin Khan, Ibrahim Elbatal, Norma Latif Fitriyani, Yeonghyeon Gu, Muhammad Syafrudin

Institutions: HITEC University, King Saud University, Imam Mohammad ibn Saud Islamic University, King Salman Center for Disability Research, Sejong University

Language: English

Topic: Hand Gesture Recognition Systems

Subfield: Human-Computer Interaction

Field: Computer Science

Domain: Physical Sciences

SDG: Reduced inequalities

Open Access status: gold

Estimated APC paid: $1,395