An Affordable Intelligent Navigation Backpack for the Visually Impaired: Deep Learning-Based Obstacle Detection and Real-Time Navigation with RGB-D Integration

Abstract

According to the World Health Organization (WHO), approximately 285 million people worldwide suffer from some form of visual impairment, including 39 million who are blind and an additional 246 million experiencing severe visual impairment. Existing navigation aids often fail to provide a user-centric perspective, relying on secondary judgment and leading to inconvenience. High-tech devices, such as smart glasses and robots, offer more effective solutions but are frequently cost-prohibitive. This study presents an affordable, first-person perspective intelligent navigation backpack for the visually impaired, utilizing deep learning. The system integrates RGB images and depth maps via alignment algorithms, extracts obstacle contours through binary image processing, and detects obstacles in real-time using the YOLO model. Experimental results demonstrate that the navigation depth camera significantly outperforms traditional ultrasonic and LiDAR sensors, achieving up to 98% measurement accuracy.