Gesture to Meaning: A Deep Dive into Video Sign Language Recognition

Abstract

 Automated systems that can recognize and understand sign languages are essential for fostering accessibility and inclusion for the Deaf and Hard of Hearing community. Unfortunately, this community still faces considerable barriers when it comes to communication, education, employment, and full participation in society. Traditional methods for sign language recognition—like rule-based systems or basic machine learning models—often fall short, as they struggle to capture the nuanced and dynamic essence of sign language. Enter Advanced Convolutional Neural Networks (CNNs), which offer a promising way forward by effectively modeling long-range dependencies and temporal sequences in communication. In recent years, innovative deep learning architectures—such as CNNs, Recurrent Neural Networks (RNNs), Multi-Task CNNs (MTCNNs), and transformers—have started to reshape the landscape of sign language recognition. This paper aims to delve into these approaches, comparing them across various metrics like accuracy, dataset size, architecture, and training time. By identifying gaps and potential improvements in video sign language recognition, we hope to enhance these systems and, ultimately, empower the Deaf and Hard of Hearing community to communicate more freely and effectively.