Hand Gesture Recognition using Depth Data for Indian Sign Language

Abstract

It is hard for most people who are not familiar with a sign language to communicate without an interpreter. Thus, a system that transcribes symbols in sign languages into plain text can help with real-time communication, and it may also provide interactive training for people to learn a sign language. A sign language uses manual communication and body language to convey meaning. The depth data for five different gestures corresponding to alphabets Y, V, L, S, I was obtained from online database. Each segmented gesture is represented by its timeseries curve and feature vector is extracted from it. To recognise the class of input noisy hand shape, distance metric for hand dissimilarity measure, called Finger-Earth Mover’s Distance (FEMD) is used. As it only matches fingers while not the complete hand shape, it can distinguish hand gestures of slight differences better.