Recognition of Indian Sign Languageusing SURF, BoW & CNN

Communication is indispensable for living in this world. However, Deaf and Mute people struggle while communicating with others. Sign Language (SL) allows such people to communicate with other members of society. Hand gesture-based communication has applications in different fields, including auto UIs, wellbeing, clinical frameworks, emergency, calamity help, and human-robot/PC collaboration. To recognize ISL alphabets and integers and convert them into text, we create an ISL dataset with 35 classes, each with 1200 images. In this paper, the Canny Edge Detection Technique is used to pre- process these images, Speeded Up Robust Feature (SURF) is used for feature extraction, and K-means clustering is used to cluster these features. Visual word vectors and classification are obtained using the Bag of Visual Words and CNN, respectively. The proposed system's reliability is assessed using different machine learning classifiers like KNN, SVM, & LogisticRegression for the same dataset which provides an average accuracy of 99.3% in the case of CNN, which is significantly higher than numerous earlier approaches.