R-DCNN Based Automatic Recognition of Indian Sign Language

Abstract

AbstractAn implementation of a system that can recognize the sign language in real time would be highly beneficial to the deaf and dumb community. The proposed work aims to develop a system for automatic recognition of hand gestures pertaining to finger spelling in the Indian sign language System (ISL) and thereby overcome the barrier that the disabled faces to communicate. The proposed R-DCNN automatic recognition is the combination of Region of Interest (ROI) and deep convolution neural network (DCNN) layers. Dataset is created with 26 alphabets and 9 numbers of 1200 gestures for each and hence 42,000 gestures are trained. The system proposes to use Three Conv2d layers followed by one dense as the hidden layer. Finally, one dense layer containing 35 nodes is used for recognizing the gesture. The time consuming for training in CNN is overcome by applying Region of Interest algorithm. Hence, the combination of neural networks gives highly efficient and accurate system for sign language recognition in real time.KeywordsIndian Sign Language (ISL)Region of Interest - Deep Convolutional Neural Network (R-DCNN)Hand Gesture Recognition (HGR)Convolutional Neural Network (CNN)Deep Neural Network