EUVIMEDEUVIMED entdecken
Crossref

DUAL-REPRESENTATION DEEP LEARNING FRAMEWORK FOR REAL-TIME SIGN LANGUAGE TRANSLATION USING GRAPH SKELETON MODELING

P. Anbumani, S. Geetha, S. Prabakaran, Kavin Kumar S, S. Gunasekaran, Vishwa M

International Journal of Drug Delivery Technology · 01.07.2026 · 16 (60s)

DOI: 10.25258/ijddt.16.60s.149 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 01.07.2026. Verfasst von P. Anbumani und 5 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

The translation systems in sign language are slowly becoming important assistive technologies needed to facilitate communication between the hearing-impaired and the hearing community. Although recent methods made in deep learning, most of the current gesture recognition systems are based predominantly on convolutional network architectures, which look at the visual appearance of gesture frames without considering the structural relationship between finger joints and the contextual information available in gesture frames.

Objective

In order to overcome these weaknesses, this paper presents a dual-representation deep learning model of real-time sign language translators based on Graph Convolutional Networks (GCN) and the use of Vision Transformers (ViT). Materials and Methods Under the proposed system, MediaPipe and OpenCV are used to extract hand landmarks on real-time video streams to give the accurate spatial location of the finger joints. These landmarks are represented as a skeletal graph in which joints are considered the nodes and anatomy provides edges. The GCN module is able to learn the motion pattern and the spatial relationship among the joints and with this, it is able to represent the articulation of the hand correctly. Meanwhile, the Vision Transformer works on gestures pictures by splitting them into image patches and using self-attention mechanisms to gather information about the entire context. The features produced by the two models are merged with the help of feature fusion module to enhance the performance of gesture classification. The identified gestures are then converted into the text and a text to speech interface.

Results

Empirical evidence shows that the suggested framework increases the recognition accuracy and creates an effective real-time communication within assistive applications.

Zitieren

APA

Anbumani, P., Geetha, S., Prabakaran, S., S, K. K., Gunasekaran, S., & M, V. (2026). DUAL-REPRESENTATION DEEP LEARNING FRAMEWORK FOR REAL-TIME SIGN LANGUAGE TRANSLATION USING GRAPH SKELETON MODELING. International Journal of Drug Delivery Technology, 16(60s). https://doi.org/10.25258/ijddt.16.60s.149

Vancouver

Anbumani P, Geetha S, Prabakaran S, S KK, Gunasekaran S, M V. DUAL-REPRESENTATION DEEP LEARNING FRAMEWORK FOR REAL-TIME SIGN LANGUAGE TRANSLATION USING GRAPH SKELETON MODELING. International Journal of Drug Delivery Technology. 2026;16(60s). doi:10.25258/ijddt.16.60s.149

BibTeX
@article{Anbumani2026,
  author = {Anbumani, P. and Geetha, S. and Prabakaran, S. and S, Kavin Kumar and Gunasekaran, S. and M, Vishwa},
  title = {DUAL-REPRESENTATION DEEP LEARNING FRAMEWORK FOR REAL-TIME SIGN LANGUAGE TRANSLATION USING GRAPH SKELETON MODELING},
  journal = {International Journal of Drug Delivery Technology},
  year = {2026},
  volume = {16},
  number = {60s},
  doi = {10.25258/ijddt.16.60s.149}
}
Kostenlos starten →