Breaking the Silence: Convolutional Neural Networks for Sign Language Recognition in the Deaf Community

Abstract

The deaf community faces communication barriers that hinder their interaction with the hearing world. This work aims to bridge the gap by enabling accurate recognition of Arabic sign language gestures. The proposed Convolutional Neural Networks (CNNs) architecture is designed to effectively capture the spatial features inherent in sign language gestures, thereby enhancing recognition accuracy. A distinctive aspect of our work involves the integration of a CNN architecture with a residual design, which effectively captures intricate spatial features inherent in sign language gestures, thereby enhancing recognition precision. The study leverages the ArSL2018 dataset, a comprehensive collection of grayscale sign language images with diverse lighting conditions and backgrounds. Our custom-built CNN model is trained on this dataset, utilizing a specialized learning rate scheduler for improved convergence. The experimental results showcase promising performance, demonstrating the potential of CNNs in sign language recognition. Furthermore, we present visualizations of the model's predictions using t-SNE, revealing the clustering patterns of different sign language gestures.