NavSense: AI-Powered Smart Glasses for Real-Time Object and Obstacle Detection for Visually Impaired and Low Vision People

Abstract

One out of 2.2 billion of the world's population lives with vision loss, though Electronic Travel Aides (ETAs) are affordable, yet appear to be only accessible in low-income regions. NavSense is a smart-glass wearable AI-based system that aims to address three major constraints of the currently available ETAs: a long processing latency, reliance on the cloud, and its prohibitive nature. Computation All computation is done on an Allwinner A733 Single Board Computer with INT8-quantized YOLOv8-nano object detection, MiDaS monocular depth estimation, and a two-sensor ultrasonic array. Hazard Priority Scoring (HPS) function interoperates semantic classification and proximity and closure rate to render only vital audio alerts. Campus-wide experiments using 85.2% mAP at 0.5 indoors and 78.4% outdoors with an end-to-end latency of 135 ms - maintained 6.81 m of reaction distance at the normal walking pace. Having a bill-of-materials under 250, NavSense proves that the technical and economic viability of high-performance, privacy-sensitive ETA hardware has become a reality.