CNN based Approach for Sign Recognition in the Indian Sign language
Pranav Unkule, Chatak Shinde, Pratik Saurkar, Sanchit Agarkar, Usha P. Verma · 2022 International Conference on Augmented Intelligence and Sustainable Systems (ICAISS) · 2022
Vocal communication is a way to express thoughts, opinions, and emotions. There are some special people who find it difficult to communicate vocally as they suffer from Nonverbal Autism. Communication between people with speech impairment and non-signers becomes difficult as they tend to use Sign Language. Each individual is not able to understand Sign Language. In this paper, a model is proposed which converts the gestures into text and furthermore to audio format. The main aim is to bridge the communication gap between speech impairment people and non-signers. By taking video input from the camera, AlexNet, which is a Convolution Neural Network architecture, can classify real-time hand gestures. During training, the model has been trained on more than 400 images of 10 gestures consisting of alphabets, numbers and phrases, which are then preprocessed with canny edge filtering in order to improve accuracy. Along with it, the recognized gesture is converted into audio. The model has achieved an accuracy of 95.31 % with a 90:10 training-testing ratio.