Real-Time Sign Language Recognition using ConvNeXt

Real-Time Sign Language Recognition using ConvNeXt aims to fill the communication gap for speech and hearing-impaired individuals by converting hand movements into readable text with a light deep-learning mechanism. Conventional sign language recognition systems are plagued with high computational expense, light sensitivity, and non-adaptability to varied hand shapes and positions. To get rid of these issues, this project uses ConvNeXt, a convolutional architecture for feature extraction and real-time processing. The model is trained on a hand sign dataset, that uses preprocessing methods like cropping, normalization, and data augmentation to increase robustness. Experimental results show a 96% accuracy rate that is better than conventional methods without sacrificing efficiency on resource-limited devices. This work shows the promise of ConvNeXt in assistive communication technologies with future work to be focused on increasing the set of gestures and improving deployment on mobile devices.