Recognition of Bengali Sign Language using Novel Deep Convolutional Neural Network
Md. Jahangir Hossein, Md. Sabbir Ejaz
Abstract
On our planet, speech and hearing-impaired people are a part of our society. When an interaction is needed between the impaired and the general people, communication becomes difficult. In several races, impaired people practice various sign languages for communication. For speech and hearing-impaired people, sign language is the fundamental communication method in their lifestyle. However, it is incredibly challenging to desegregate them into the mainstream because the majority part of our community is not aware of their practiced sign language. Nowadays, computer vision-based solutions remain fully appreciated to get their sign language comprehensible to general people. Many analysts are taking a shot at Recognition of Hand Gesture, one of the computer vision-based solutions to recognize sign language. It's been a popular area for research for an extended period now. Some recent studies have reached immense performance using models of deep learning in the region regarding Hand Gesture Recognition in Computer vision. Through this research work, our aim to reduce the communication difficulties among the speech and hearing-impaired people and the rest of Bangladesh by building an appropriate deep learning model that can recognize Bangla Sign Language alphabets precisely. In this work, a different CNN (Convolutional Neural Network) architecture is introduced to identify the alphabets of Bengali sign with the respective Ishara-Lipi database. This architecture accomplished a general precision of 99.86%, which surpassed all prior works regarding Bengali sign alphabet recognition.