Adapting YOLO to recognition of real numbers in 7-segment digits

Detection of real numbers represented as 7-segment digits is usual in applications involving the automatic detection of measurements in labs. However, the performance of most extended Optical Character Recognition (OCR) tools for this specific task is quite poor. In this work, we transform the very well known YOLO object detection algorithm to an OCR algorithm able to detect and recognize real numbers represented by 7-segment digits. A set of images labeled in YOLO format is first needed in order to retrain a specific YOLO model. Then, the re-trained YOLO inference output is used to translate the images representing digits of a real number. The results obtained show that the performed translations have a high degree of confidence but there is a possibility of confusing digits or detecting more than one decimal point that can be later corrected.