Authors

Jiuling Zhong, Yifei Wang

Research Question

Can a computer vision-enabled smart wearable system effectively assist PBLV in navigating construction sites?

Background

Navigating urban construction zones poses significant challenges for People with Blindness and Low Vision (PBLV), with obstacles like scaffolding and barriers, and warning signs that are often undetected by existing tools. Assistive technologies lack real-time construction site awareness, increasing safety risks. This project develops a wearable computer vision-powered system using YOLO object detection to identify hazards, Optical Character Recognition (OCR) to interpret warning signage, and motion analysis to assess the impact of movement on detection accuracy.

Methodology

First, a database was built. Photos were captured of various construction sites at different angles and in different weather conditions to enrich the database. At the same time, the code was optimized to increase the accuracy of construction site recognition and remove non-target recognition. The next step was for the prototype device to recognize text while the blind users were walking. Four OCR engines were benchmarked—Google OCR, EasyOCR, Tesseract OCR, and PaddleOCR—to evaluate their text detection accuracy on actual construction sites. Different camera configurations—glasses, chest-mounted, and handheld—were also evaluated to increase the accuracy of PaddleOCR recognition and improve the database. Next, a motion impact analysis was performed and walking speeds (2m/s, 3m/s, 4m/s, 5m/s) were tested to measure the stability of the system and the accuracy of PaddleOCR in motion.

Deliverables

• Functional Prototype of a real-time computer vision wearable device for navigation assistance
• Technical Report

Datasets