Partial Hand Sign Recognition and Sentence Construction for Diverse Age Groups using VGG19 and Seq2Seq

The development of a sign language recognition (SLR) system with the goal of closing the communication gap between deaf and hearing people. The system supports several languages and makes age-appropriate adjustments, allowing sign language users of various linguistic backgrounds and ages to communicate effectively. The study details the creation of a huge sign language dataset, as well as the deployment of machine learning models and recognition algorithms for exact sign identification and translation. The VGG19 model, trained on this dataset, achieved a precision rate of 99.90% in recognizing English sign words. The system architecture includes an encoder-decoder setup with a dense layer, and a sequence-to-sequence (Seq2Seq) translation model for translating recognized English sign words into Indian regional languages. The results demonstrate remarkable accuracy in SLR gestures across linguistic boundaries, with the Seq2Seq model enabling seamless translation, making sign language communication more inclusive. A rolling average technique is used to enhance the interpretability of sign language videos. The research sets the stage for future enhancements in language support, video classification, and user interface refinements, contributing to a world where sign language is universally understood and accessible.