Abstract:
Through the creation of a Convolutional Neural Network (CNN)-based model that is
inspired by the YOLO architecture, the purpose of this thesis is to find a solution to the
substantial difficulty of Braille character recognition. The purpose of this research is to
prioritize accessibility for those who are visually challenged and rely on Braille for
communication and education. Specifically, the research is designed to transform
Bangla letters from Braille patterns. For the purpose of ensuring that the training and
testing stages are robust, a dataset that was painstakingly selected and specially
designed for this particular task was painstakingly collected and annotated. However,
the CNN model had difficulty reliably identifying contextual characters within
sentences or phrases, despite the fact that it was able to recognize individual Bangla
characters within Braille patterns with exceptional competence. The study highlights
the important need to improve the accuracy and durability of such models in order to
attain practical utility in applications that are based in the real world. For the purpose
of model training, the system makes use of Google Colab Pro. Additionally, it makes
use of advanced GPU capabilities and makes use of TensorFlow and Keras libraries for
efficient implementation. In the future, efforts will be focused on refining the model in
order to increase contextual character identification. The ultimate objective is to
broaden the range of educational options available to visually impaired people and to
improve the quality of life for those individuals who rely on Braille as an essential form
of communication and literacy support. By tackling these problems, our research makes
a contribution to the advancement of accessible technology and aids the inclusion of
visually impaired individuals in educational settings and in day-to-day life.
