DART.
AI-powered tunnel inspection from images. Detect damage type, severity, and extent — fast analysis, interactive visualizations, clear insights.
What it does.
DART automates the inspection of tunnels and similar infrastructure using computer vision. It turns raw image data into actionable damage reports without manual surveying.
Automated detection
Convolutional neural networks trained on labeled inspection datasets classify cracks, spalling, corrosion, and water ingress from standard camera images.
Severity mapping
Each detected defect is assigned a severity grade and a spatial location, producing a damage map of the entire tunnel segment.
Interactive reports
Results are presented in a browser-based viewer where engineers can zoom into individual defects, filter by type, and export structured data.
How it works.
The pipeline runs from image capture to structured output in a single pass.
Image acquisition
Standard camera images collected during routine tunnel inspections. No special lighting or positioning required.
Feature extraction
A deep convolutional backbone extracts visual features at multiple scales, capturing both fine cracks and large-area damage.
Classification and localization
Defects are classified by type and severity, and localized in the image coordinates and mapped to the tunnel geometry.
Reporting and export
An interactive viewer aggregates the results into a per-segment damage summary, with export to standard asset-management formats.
Where it applies
- Rail and road tunnel structural health monitoring
- Bridge underside and deck inspection
- Pipeline interior survey
- Historic building facade assessment
- Post-disaster rapid damage evaluation
Limits of this model
- Detection accuracy depends on image quality and lighting conditions
- Model trained primarily on concrete tunnels — masonry and steel require additional data
- Real-time processing requires GPU acceleration for large image volumes
- Structural engineering judgment still required for final classification