Publication

Abstract : According to a 2018 report by the World Health Organization, nearly 63 million people in India suffer from partial or complete hearing impairment. They communicate using sign language (SL), a language that uses a set of specific non-verbal gestures, mainly involving hands. One major barrier in their way of communication is that the majority of the non-hearing and speech impaired population does not understand sign language, which raises the necessity of developing sign language recognition systems that can be standardized across the nation. This paper aims at giving the best recognition model for Indian Sign Language. The system involves the Bag of Visual Words technique for real-time prediction of the ISL with a comparative study of feature detectors and descriptors like SIFT, SURF, ORB, STAR + BRIEF, and FAST + FREAK. The results indicate that using the SURF feature detector and descriptor along with SVM yields an accuracy of 99.94% and gives a rotation threshold of up to 5 degrees. Meanwhile, parallel experimentation and comparison with the concept of CNN and ASL dataset have shown promising results such as 100% accuracy for ISL with CNN, 65.41% accuracy for ASL with SURF and SVM, and 99.77% accuracy for ASL with CNN.