Vision-Based Tunnel Lining Health Monitoring via Bi-Temporal Image Comparison and Decision-Level Fusion of Change...

Area of Science:

• Civil Engineering
• Computer Science
• Robotics

Background:

• Traditional tunnel structural health inspections rely on subjective, time-consuming visual observations.
• On-site inspections pose risks to human surveyors and can necessitate operational shutdowns.
• There is a need for automated, accurate monitoring systems to improve tunnel structural health assessment.

Purpose of the Study:

• To develop and evaluate a remotely operated machine vision change detection application for enhanced tunnel structural health monitoring.
• To mitigate the limitations of manual inspections, including subjectivity, time consumption, and safety hazards.

Main Methods:

• A vision-based sensing system using a robotic platform with cameras to capture tunnel wall data.
• Image processing and deep learning for pre-processing to minimize light variation effects.
• Image fusion techniques and pixel-based change detection to identify structural changes.
• Decision-level fusion to combine change maps for reliable detection of alterations between inspections.

Main Results:

• The proposed system achieved a recall of 81%, precision of 93%, and an F1-score of 86.7% in quantitative analysis.
• Demonstrated reliable detection of changes occurring in tunnel structures between surveys.
• Successfully reduced nuisance changes caused by lighting variations through advanced image processing.

Conclusions:

• The developed machine vision system offers a more accurate and reliable method for tunnel structural health monitoring.
• Remote operation and automated analysis enhance inspection efficiency and surveyor safety.
• The system provides a robust solution for detecting structural changes, improving overall tunnel maintenance.