EUVIMEDEUVIMED entdecken
Crossref

Enhancement of Recurrent Neural Networks (RNN) applied in hand gesture recognition for American Sign Language (ASL) alphabet recognition

Jenny Jimenez, Augustin Brain Sabordio

International Student Research Review · 30.04.2026 · 3 (1) · 22

DOI: 10.53378/isrr.212 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Student Research Review, online erschienen am 30.04.2026. Verfasst von Jenny Jimenez und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study focused on improving Recurrent Neural Network (RNN) algorithms to recognize American Sign Language (ASL) alphabets to better hand segmentation, feature extraction, and time modeling to obtain better accuracy and robustness in the real-time recognition. An experimental research design was employed using 260 video samples representing all 26 ASL letters under both ideal and challenging environmental conditions. The enhanced model integrates MediaPipe-based hand detection with adaptive preprocessing, multimodal feature extraction combining 3D landmarks and engineered articulation features (99-dimensional vectors), and adaptive temporal modeling using extended sequence buffering and prediction smoothing. Dual-stream neural architecture takes visual and numerical data of landmarks and processes them before being classified through LSTM layers and softmax output. The improved system achieved an overall accuracy of 97.70% and a mean confidence of 91.45%, which is 38.85 percentage points higher than the baseline model. Accuracy in challenging conditions was significantly improved, with a degradation rate of only 1.60% compared to 8.50% in the baseline. The recognition of visually similar letters reached 98% accuracy, while dynamic letters J and Z achieved relative improvements of 160% and 126%, respectively. The current research study is limited to the recognition of ASL alphabets (A–Z) in controlled experimental conditions. Future research may extend the system to full-word recognition and real-world deployment scenarios.

Zitieren

APA

Jimenez, J., & Sabordio, A. B. (2026). Enhancement of Recurrent Neural Networks (RNN) applied in hand gesture recognition for American Sign Language (ASL) alphabet recognition. International Student Research Review, 3(1), 22. https://doi.org/10.53378/isrr.212

Vancouver

Jimenez J, Sabordio AB. Enhancement of Recurrent Neural Networks (RNN) applied in hand gesture recognition for American Sign Language (ASL) alphabet recognition. International Student Research Review. 2026;3(1):22. doi:10.53378/isrr.212

BibTeX
@article{Jimenez2026,
  author = {Jimenez, Jenny and Sabordio, Augustin Brain},
  title = {Enhancement of Recurrent Neural Networks (RNN) applied in hand gesture recognition for American Sign Language (ASL) alphabet recognition},
  journal = {International Student Research Review},
  year = {2026},
  volume = {3},
  number = {1},
  pages = {22},
  doi = {10.53378/isrr.212}
}
Kostenlos starten →