Publication

Abstract :

Sign language recognition can provide excellent platform for the hearing-impaired people in society. In day-to-day life physically challenged people face problem to communicate with normal people. There are many research works done for sign language recognition, but there is, lack of availability of dataset for the research work which leads to less development of tools for training and testing for the recognition of signs. This paper proposes an algorithm that takes offline NEWS video stream to create dataset of sign language video with the corresponding text. The signs are recognized from the gesture of the person in the news-video with the help of lip movement, facial expression. Therefore, those signs will be mapped to the audio text in the video up to sentence level. The proposed algorithm takes input of offline video and apply various techniques such as extraction, segmentation, template matching, speech conversion to generate dataset which consist of sentence level segmented videos which will be pushed to pre-trained model to get the output for the sign language which will be compared with the audio text and comparison score will be calculated. Therefore, this technique will help to enhance the dataset for the future research for Sign Language recognition.