Enhancing Taiwanese Sign Language Recognition by Using Data Augmentation and Deep Learning

Taiwanese Sign Language (TSL) recognition is crucial for communication accessibility. However, the lack of large-scale TSL datasets limits deep learning models' generalization. To address this, we propose a data augmentation strategy to enhance dataset diversity and model robustness. Our approach integrates various augmentation techniques, such as rotation, flipping, and brightness adjustment, to generate diverse training samples while preserving key gestures. We employ the MoViNet architecture, optimized for sequential action recognition, to train our model. Experiments confirm our approach improves accuracy, especially in low-resource settings. This study highlights the importance of dataset expansion in sign language research and demonstrates the effectiveness of deep learning in improving TSL recognition.