Reconstruction of Convolutional Neural Network for Sign Language Recognition
Rahib H. Abiyev, John Bush Idoko, Murat Arslan
Abstract
This paper presents a Sign Language translation model using Convolutional Neural Networks (CNN). A sign language is a language which allows mute and hearing-impaired people to communicate. It is a visually oriented, nonverbal communication which facilitates communication through body/facial postures, expressions and a collection of gestures. To contribute to the wellbeing of the affected population, we are motivated to implement a vision-based system to avert their day to day challenges. Our propose model constitutes object detection and classification phases. The first module is made up of single shot multi-box detector (SSD) used for hand detection. The second module constitutes convolutional neural network plus a fully connected network utilized to constructively translate the detected signs into text. The propose model is implemented using American sign language fingerspelling dataset. The propose system outperformed other published results in the comparative analysis, hence recommended for further exploitation in sign language recognition problems.