AESRM-Automatic English Sign Language Recognition with Machine learning techniques

The hand sign recognition has one of the most important learning domains and applications in the fields of computer vision and artificial intelligence. More specifically, one type of such communication is Sign Language, which includes finger gestures, illustrations of the face and gestural movements. However, there is always a social barrier that separates the deaf community from the oral/aural society, and hence there is need to develop better and more natural channels of communication. Making a stand that is one of the potential remedies is capable of correctly identifying the performed hand gestures and translating them into corresponding letters of American Sign Language (ASL) almost immediately. For image collection in this paper, first the images are collected from a live video stream and then these images are preprocessed to form the dataset. Then, the given dataset is divided into testing and training portions set over which, various machine learning algorithms like Naive Bayes, Multinomial Naive Bayes, Random Forest, SVM, KNN, Logistic Regression, Decision Trees are built, and a model/classifier and highest accuracy of 99. 79% has been achieved by decision tree. Lastly, to identify the hand signs in the live video stream we use the predicted hand signs, and a string is produced which can be converted into the user's preferred speech.