Helping Blind Individuals: Pakistani Currency Recognition System for Blind People Using the YOLOv8 Model

Abstract

Blind or visually impaired individuals face significant challenges in recognizing currency notes, especially in regions where cash transactions remain common. To address this issue, we propose a real-time currency detection system tailored for Pakistani currency notes. The system is powered by the YOLOv8 object detection model, which achieved 99.1% accuracy with an average inference time of 12 milliseconds per image, ensuring both precision and speed. Once the denomination is identified, it is communicated to the user via text-to-speech for audible feedback. The application is structured with a Streamlit-based front-end for user interaction and a Flask-based back-end API, deployed via NGROK to ensure secure, accessible cloud usage. This architecture allows real-time operation across devices without local installations. In addition to implementation, we conducted a comparative evaluation of YOLOv8 against VGG19 (val-accuracy 83%), highlighting YOLOv8's superior performance in terms of accuracy and inference speed across varied lighting and orientation conditions. The results reinforce YOLOv8's suitability for assistive technologies and provide a benchmark for future improvements in AI-powered accessibility tools.