A Comprehensive Study on Relative Distances of Hand Landmarks Approach for American Sign Language Gesture

Communication with people with hearing or speaking disabilities is always difficult when there is no knowledge of sign language. The presence of sign language is not enough to communicate smoothly, this process requires another easy medium for communication to make it more efficient, that is, via a digital medium. This paper proposes using Feed-Forward Neural Networks on hand landmarks for real-time sign language identification. The hand landmarks identification was carried out using the MediaPipe Hands library. This approach would make the classification problem efficient by making it faster and requiring less memory. Through this, we aim to bridge the gap between the difficulties that arise during communication between people who do and do not know American Sign Language.