Image Translation of Bangla and English Sign Language to Written Language using Convolutional Neural Network

Abstract

One particular thing that differentiates humans from other species is their abilities to interact. to communicate with others, humans invented languages as units. There are 6500 languages in this world and English has been established as a global language. However, there are a ton of physically disabled human beings who are deprived of expressing their emotions through verbal language. Therefore, sign language has been discovered by expressing feelings with the help of signs which is mainly done by moving body parts: hands in particular. Although, it is so rare to find a research where both Bangladeshi Sign Language (BdSL) and American Sign Language (ASL) is translated, previously some of the prominent researchers worked on primary or secondary ASL and BdSL datasets separately and obtained high accuracy (> 97%) based on their algorithmic approaches. The executed system focuses on implementing both of the aforementioned sign languages combinedly as well as working on both the primary and secondary datasets using single algorithmic approach in order to resolve the two way communication and a better understanding of communicating through sign language. In this thesis, the advantages of real world pictures of Bangladeshi Sign Languages will be used to run an algorithm which will convert sign language to written language using Sequential Convolutional Neural Network method. The system will be able to detect both BdSL and ASL regarding any background with the accuracy of 95.23% and 98.45% respectively.