YOLO-v3 Based Currency Detection and Recognition System for Visually Impaired Persons
Rakesh Chandra Joshi, Saumya Yadav, Malay Kishore Dutta
Abstract
It is easy for a normal human being to perceive and recognize any banknote effortlessly but it is really much difficult for any visually impaired or blind person to perform the same task. As money has an important role in daily lives for any business transaction, real-time detection and recognition of banknotes become a necessity for a person especially who is blind or visually impaired. For that purpose, the YOLO-v3 CNN model based banknote detection and recognition system is proposed which is fast and accurate. Images of different denominations and in different conditions were are collected initially and then, these images are augmented with different geometric and image transformations on images, to make the system robust. These augmented images are then annotated manually, from which training sets and validation image sets are prepared. Later, the performance of the trained model has evaluated on a real-time scene as well as a test dataset. The test result shows that the proposed YOLO-v3 model based method has detection and recognition accuracy of 95.71% and 100%, respectively. The whole system is standalone and works in real-time.