Bangla Sign Language Recognition Using Concatenated BdSL Network

Abstract

Communication has always been a challenge for the deaf-mute community. So sign language is the only way of interaction for them. But the problem is that sign language is way too complex for the general mass. Keeping this in mind we propose an effective alternative tool to recognise Bangla Sign Language (BdSL) using computer vision for the people in Bangladesh. In our research we propose a novel architecture, namely "Concatenated BdSL Network" combining Convolutional Neural Network (CNN) as an "Image Network" for visual feature extraction and a pretrained "Pose Estimation Network" for extraction of the hand keypoints from hand gestures. This research will hold promising future aspects for real-time sign language interpretation.