Machine Learning Based Smart System for Blind and Visually Impaired People

The World Health Organization estimates that 39 million people worldwide are totally blind, out of 284 million with visual impairment. This paper presents a comprehensive methodology and evaluation of a smart system for blind and visually impaired individuals. The system integrates machine learning modules including a Personal Assistant Bot (PAB), Face Recognition, Object Detection using YOLOv7, Optical Character Recognition (OCR), Distance Measurement and Depth Estimation, Navigation, and Bangladeshi Bank Note Detection. The methodology outlines the system's architecture and operational framework, emphasizing real-time processing and user accessibility. Performance evaluations highlight high accuracy rates in Face Recognition (99%), Object Detection (95.7% precision), and Distance Measurement (96.33%). Usability is validated through real-world tests, demonstrating its potential to enhance the autonomy and safety of visually impaired users.