Empowering the Blind School Communities by Recognition of Handwritten Braille Text Documents Using YOLOv5 Technique

Braille is a tactile writing system that consists of raised dots arranged in a 3×2 grid for those who are visually handicapped. To empower blind school populations, this project focuses on identifying handwritten Kannada Braille text documents specifically gathered from educational institutions. The handwritten Braille has unique challenges such as variability in dot formation, differences in spacing of dots, and limited available datasets, which impact recognition accuracy and reliability. Recent findings in deep learning such as Convolution Neural Networks [CNNs] and You Only Look Once [YOLOv5]-based models have improved image recognition but struggle with irregularities in handwritten Braille texts. To address these challenges, this project aims to empower blind schools by accurately identifying the Kannada handwritten Braille texts using YOLOv5, a real-time object detection. A dataset of 1500 samples from Kaggle is used to train the model, and 1000 samples gathered from these educational institutions are used to validate it. With a 5.832% loss and an accuracy of 94.168%, the results show that YOLOv5 is a useful technique for real-world Kannada Braille character recognition.