Real-Time Static Hand Sign Recognition System using MediaPipe and Fully Connected Neural Network

About 5% of the world's population faces the challenges of being deaf and mute, according to the World Health Organization. This highlights the urgent need for communication technologies to connect individuals with speech and hearing impairments to the wider community, where Sign Language Recognition (SLR) is a pioneer. However, the development of sign language recognition systems is often challenged by the scarce availability of suitable datasets, as each spoken language has its own distinct sign language. Additionally, the high computational power required to process this data complicates the progress, making it hard to reach the masses. This paper presents a real-time static sign recognition system that not only creates robust and customizable datasets but also preprocesses the data, and recognizes static hand signs, making it adaptable to any static sign language. Currently based off the American Sign Language (ASL), the model uses MediaPipe and a fully connected neural network (FCNN) that is trained and tested on 10 signs with 200 images per sign and later with 500 images per sign to include diversity. The proposed system demonstrates high accuracy and fast recognition, offering a practical solution for enhancing communication for speech and hearing-impaired individuals in real time.