DESIGN AND IMPLEMENT A LOW-COST, WEARABLE VISUAL AID FOR VISUALLY IMPAIRED INDIVIDUALS
Faculty
Year of Publication
Publication Type
Abstract
Visual impairment significantly restricts independent navigation and environmental awareness, affecting over 2.2 billion people globally. This project aims to design and implement a low-cost, wearable visual aid that enhances navigation safety for visually impaired individuals through real-time object detection and intelligent audio feedback. The system addresses critical gaps in affordability, offline capability, and intelligent information prioritization by providing a practical alternative to expensive commercial assistive devices.
The system integrates a Microsoft LifeCam HD-3000 camera module with a Raspberry Pi 5 single-board computer running the YOLO-Fastest object detection model via the NCNN inference framework. It encompasses a four-stage pre-processing pipeline including frame resizing, color conversion, normalization, and data layout transformation. A spatial processing and prioritization algorithm evaluates detected objects based on monocular distance estimation, directional zoning, trajectory analysis, and hazard level classification. The system generates context-aware audio feedback through the pyttsx3 text-to-speech engine, employing priority-based speech rate modulation (150-180 wpm) and an interrupt mechanism for critical warnings.
Comprehensive testing was conducted across five detection scenarios, range validation tests, object prioritization assessments, and environmental stress tests under varying lighting conditions. Experimental results demonstrated consistent real-time performance with average response times of 55.68–65.18 ms per frame (15–18 FPS) and confidence scores ranging from 0.68 to 0.85. The system achieved effective detection ranges of 8 m for pedestrians, 33.21 m for static obstacles, and 112.9 m for vehicles. Object prioritization accuracy reached 100% in simple scenarios but decreased to 73.8% in highly crowded environments with seven simultaneous objects. Environmental testing revealed detection rates of 95–100% under daylight conditions, degrading
to 30% in near-dark environments. The system processed over 15,000 frames without software crashes, confirming operational stability. The prototype, assembled from off-the-shelf components at a fraction of commercial costs, successfully demonstrates the feasibility of an affordable, edge-deployed wearable visual aid suitable for daytime navigation assistance.
The system integrates a Microsoft LifeCam HD-3000 camera module with a Raspberry Pi 5 single-board computer running the YOLO-Fastest object detection model via the NCNN inference framework. It encompasses a four-stage pre-processing pipeline including frame resizing, color conversion, normalization, and data layout transformation. A spatial processing and prioritization algorithm evaluates detected objects based on monocular distance estimation, directional zoning, trajectory analysis, and hazard level classification. The system generates context-aware audio feedback through the pyttsx3 text-to-speech engine, employing priority-based speech rate modulation (150-180 wpm) and an interrupt mechanism for critical warnings.
Comprehensive testing was conducted across five detection scenarios, range validation tests, object prioritization assessments, and environmental stress tests under varying lighting conditions. Experimental results demonstrated consistent real-time performance with average response times of 55.68–65.18 ms per frame (15–18 FPS) and confidence scores ranging from 0.68 to 0.85. The system achieved effective detection ranges of 8 m for pedestrians, 33.21 m for static obstacles, and 112.9 m for vehicles. Object prioritization accuracy reached 100% in simple scenarios but decreased to 73.8% in highly crowded environments with seven simultaneous objects. Environmental testing revealed detection rates of 95–100% under daylight conditions, degrading
to 30% in near-dark environments. The system processed over 15,000 frames without software crashes, confirming operational stability. The prototype, assembled from off-the-shelf components at a fraction of commercial costs, successfully demonstrates the feasibility of an affordable, edge-deployed wearable visual aid suitable for daytime navigation assistance.
Supervisor(s)
co-supervisor