Analysis of Vision based Techniques for the Translation of Indian Sign Language

Abstract

Sign language acts as a medium of communication among those of the hearing impaired and mute community. However, it cannot be easily understood by common people. Various research has been done to bridge this gap by developing Sign Language Recognition (SLR) methodologies. Studies say that 1 in every 5 deaf people is Indian.

In this paper, a thorough review of these methodologies has been done, to compare and contrast various aspects of them. This includes an overview on different preprocessing methods used like segmentation, image morphological processing, cropping, etc, feature extraction techniques like Fourier Descriptors, Image Moments, Eigen values, Mediapipe and others. This study also covered classification models spanning from Distance metrics to Kernel based approaches and feedforward neural networks, along with Deep Learning based methods such as CNNs, LSTMs, GANs, Transformers etc.