Braille Character Recognition Independent of Lighting Direction Using Object Detection Models
Akihiro Yamashita, Taichi Shirakawa, Katsushi Matsubayashi · 2024
Braille text is widely used around the world as a means of communication for the visually impaired. Although Braille is often regarded as a tool for the visually impaired, there are instances where sighted individuals may want to verify Braille documents. For example, teachers at schools for the visually impaired may need to check homeworks or assignments written in Braille, or care workers supporting the daily lives of visually impaired people may need to verify Braille documents. To meet such social demands, optical Braille recognition (OBR) technologies have typically been developed using cameras and scanners. In particular, methods using object detection models based on Deep Learning have achieved high accuracy in recent years. Since Braille is represented by embossed dots on thick paper, the appearance of shadows varies greatly depending on the angle of the lighting. Therefore, the angle and intensity of the light significantly affect the accuracy of Braille recognition. Specifically, in the case of double-sided Braille, where Braille is embossed on both sides, it is necessary to distinguish between raised and recessed dots. Previous studies have mostly imposed constraints on the angle of illumination, such as limiting the imaging method to scanners or requiring the light source to be positioned on the top of the Braille document when using a camera. However, for practical use in daily life, it is preferable to recognize Braille documents regardless of the direction and lighting conditions. In this study, we created a dataset of Braille images captured under various angle of lighting and developed a Braille recognition model that is independent of the lighting angle by fine-tuning the object recognition model. We implemented and compared RetinaNet, which has been used in previous research, and the anchor-free model YOLOX as object detection models. As a result, we achieved a model capable of detecting Braille images with an accuracy of mAP50=O.98 or higher, regardless of the angle of lighting.