Developing an Assistive Technology for Visually Impaired Persons: Ethiopian Currency Identification

Abstract

Abstract In this study an innovative assistive technology designed to empower visually-impaired individuals by providing them with the means to independently identify Ethiopian banknotes. Visually-impaired individuals often face challenges in recognizing and differentiating between currency denominations, hindering their financial independence and day-to-day transactions. To address this issue, we developed a novel device that attaches to standard eyeglasses and leverages state-of-the-art deep learning techniques for real-time banknote detection and recognition. An extensive dataset comprising over 15,000 annotated images of five Ethiopian birr denominations (5, 10, 50, 100, and 200 Birr) was used for training. YOLOv5 was selected as the best-performing model, attaining a mAP of 97.9%. Additionally, the study compared YOLOv5 with other popular object detection models such as SSD_MobileNet_v2 and Faster_RCNN_Inception_v2. YOLOv5 outperformed the other models in terms of both accuracy and speed. The research goes beyond the development of the device and extends to the user interface, introducing a mobile application named "Genzebe." This application seamlessly integrates with their device, allowing users to capture images of banknotes, which are then processed for denomination recognition. The application has audio output of denomination to the user. By making use of cost-effective hardware components, including a Raspberry Pi and a Pi Camera, the device is implemented and demonstrated as a practical solution that could significantly enhance the autonomy and financial inclusivity of visually-impaired individuals. In conclusion, this work showcases a highly effective and user-friendly solution for identification of Ethiopian currency.