Smart Sight: A YOLO Based Deep Learning System for Real-Time Instance Recognition and Assistance for the Visually Impaired

Abstract

 Visual impairment affects over 285 million people globally, significantly limiting their ability to navigate environments, recognize objects, and perform daily tasks independently. While traditional assistive technologies such as white canes and guide dogs offer limited spatial awareness, they lack the ability to provide real-time semantic understanding of the surrounding environment. This paper presents Smart Sight, a YOLO-based (You Only Look Once) deep learning system designed to provide real-time object and instance recognition for visually impaired individuals. The proposed system integrates a lightweight wearable camera module with a YOLOv8 object detection backbone, an OCR engine for text recognition, and a Natural Language Processing (NLP)-based voice output pipeline to deliver contextual audio descriptions. The platform supports obstacle detection, currency identification, scene text reading, and human recognition, all processed in real time with an average inference speed of under 15 milliseconds per frame. Experimental evaluations demonstrate a mean Average Precision (mAP) of 91.7% across all detection categories, outperforming existing CNN-based and SSD-based assistive systems. The proposed system offers a portable, low-cost, and highly accurate solution that bridges the gap between computer vision technology and accessibility for the visually impaired.