A novel model for expanding horizons in sign Language recognition

Résumé

The American Sign Language Recognition Dataset is a pivotal resource for research in visual-gestural languages for American Sign Language and Sign-Language MNIST Dataset. The dataset contains over 64,000 images meticulously labeled with the corresponding ASL gesture or letter to recognize and classify ASL signs. Recent computer vision and deep learning advances have enabled various ASL recognition techniques. However, further improvements in accuracy and robustness are still needed. This study comprehensively evaluates different ASL recognition methods and introduces a novel architecture called Sign Nevestro Densenet Attention (SNDA). All methods are evaluated on the ASL Recognition Dataset to ensure a representative evaluation. SNDA employs the Nadam optimizer for faster convergence during training. Accurate ASL classification has practical implications for improving communication accessibility. SNDA achieves state-of-the-art performance with 99.76% accuracy, perfect sensitivity, and high specificity and precision. These results validate the effectiveness of SNDA for ASL gesture recognition and highlight its potential to promote inclusivity for deaf and hard-of-hearing communities. The fused attention mechanism demonstrates how deep learning models can be enhanced for specific application domains.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0