Sign Language Finger-Spelling Recognition System Using Deep Convolutional Neural Network

Sign Language is used by deaf and hard hearing people as it is the efficient way to communicate among themselves as well as with other people. The demand for an automated system, which converts sign language into a normal message and vice versa, has been increased to provide a normal life to people with hearing difficulties. Each sign language has a distinct character with variations in the shape of the hand, the profile of movement, and the position of the hand, face, and body parts that contribute to each character. Pattern recognition and gesture recognition are majorly used technologies for resolving the problem. In this work, real-time American Sign Language is carried out for finger-spelling recognition using Deep Convolutional Neural Network for the classification of data to bridge the gap of communication between hearing disabled people and normal people. A new dataset of more than 50 thousand images is prepared for training purposes. Outputs are based on inputs given by the user, while the designed interface provides the end user with alphabets A to Z and Numerals from 0 to 9 prediction. The overall performance based on the proposed approach is being evaluated by the use of a dataset of real-depth images captured from users. Accuracy of the system is 99.70%.