earticle

영어

This paper addresses the critical need for sign language recognition to improve communication for individuals with hearing impairments. Our approach involves leveraging the capabilities of computer vision and deep learning to develop an efficient sign language recognition system. By utilizing the Mediapipe library, we extract detailed hand landmarks from video input, thereby effectively capturing the nuanced movements and configurations inherent in sign language gestures. At the core of our model is the GRU network, a type of recurrent neural network (RNN) specifically designed for analyzing sequential data. The GRU architecture excels at capturing temporal dependencies, making it a well-suited choice for recognizing the dynamic nature of sign language expressions. To train our GRU network, we leverage the LSA64 dataset, which comprises comprehensive videos featuring expert sign language users. We extract hand landmarks from these videos using the Mediapipe module, enabling our model to effectively learn and recognize the intricate patterns of sign language gestures.