Real-time Bangla Sign Language Detection using Xception Model with Augmented Dataset

Bangla Sign language (BdSL) is the communication language used by the deaf and dumb of Bangladesh. In this paper, we present an optimal approach to recognize BdSL in real-time. First, we have developed BdSLInfinite dataset, which consists of 2,000 images of 37 different signs. Using this dataset, a convolutional neural network (CNN) based model is trained using Xception architecture that achieves 98.93% accuracy over the test-set, with response time of 48.53 ms on average. To the best of our knowledge, our proposed method outperforms all existing BdSL recognition methods in terms of both accuracy and speed.