Greek sign language recognition using machine learning

This research develops a real-time Greek sign language (GSL) recognition system using machine learning and computer vision techniques to assist communication for the deaf community. The hand tracking module from the CVZone library was utilized for hand detection and tracking in video feeds. A dataset of 37,589 images capturing 24 Greek alphabet letter gestures was collected from different users under varying conditions. This dataset was used to train a deep neural network model based on the MobileNetV2 architecture. The model achieved 96.86% accuracy on the validation set during training. On the final test set, it obtained 95.66% accuracy in classifying and recognizing GSL hand gestures. Compared to prior rule-based and fuzzy clustering approaches, the model demonstrated significantly improved recognition performance. The high accuracy indicates that combining robust computer vision techniques for hand tracking with deep convolutional neural networks can be an effective approach for real-time GSL recognition. This system has the potential to facilitate communication and accessibility for the deaf and hard-of-hearing community. Further work should focus on expanding the vocabulary beyond individual letters to incorporate common words and phrases, as well as optimizing the system's real-time performance by reducing processing lags. Overall, this research demonstrates a promising machine learning and computer vision-based system for automatic GSL recognition that can aid the deaf community.