An Efficient Binarized Neural Network for Recognizing Two Hands Indian Sign Language Gestures in Real-time Environment

This paper proposes an efficient architecture based on Binarized Neural Network (BNN) for recognizing two-hand Indian Sign Language (ISL). Some of the existing works apply deep Convolutional Neural Networks (CNN) and machine learning (ML) models to solve the challenging task of ISL recognition in a real-time environment. Therefore, the proposed BNN architecture having binary weights and activations, with bitwise operations used to reduce computational complexity during training and also, the method based on skin color segmentation and Otsu thresholding is applied to extract hand region during sign recognition. Extensive experiments are carried out on the ISL database having English alphabets and words static gestures show that the proposed work achieves a higher recognition rate of 98.8% in real-time as compared to other works. Moreover, a comparative analysis of the proposed BNN is performed over CNN in terms of memory consumption, speed, and training accuracy.