Empowering Communication: AI-driven Sign Language Fingerspelling Recognition and Translation for Deaf Accessibility
Sign language recognition technology is important in ensuring inclusive communication for the Deaf and hard-of-hearing community. In this work, the researchers present a novel approach by developing a hybridized model that effectively employs Convolutional Neural Networks (CNNs) and Transformer Models for American Sign Language (ASL) recognition. The proposed model is rigorously evaluated on two established datasets, the Google American Sign Language Fingerspelling Recognition Corpus and the Google Isolated Sign Language Recognition Corpus, demonstrating superior performance with precision, recall, and F1-scores consistently around 0.93. The model achieves an overall accuracy of 94.76% on the former dataset and 98.38% on the latter, showcasing its robustness. Furthermore, the model's generalizability is proved with the help of the ChicagoFSWild+ dataset, revealing an overall accuracy of 87.59%. The research underscores the model's versatility and potential impact on real-world applications, particularly in enhancing accessibility. The findings contribute to the advancement of sign language recognition technology, emphasizing its significance in promoting inclusive communication and opening avenues for future exploration in recognizing various sign languages.