A context-aware deep learning model to assess element-wise corrosion of steel bridges
Funded by USDOT through UTC
Graduate Student: Parham Bakhtiary
This project investigates how computer vision models can translate roadway imagery into useful information for transportation asset inspection and maintenance. In collaboration with the Utah Department of Transportation, we developed a condition-assessment framework that pairs MUTCD-aligned defect annotations with convolutional (YOLOv11) and transformer-based (RT-DETR) object detectors, benchmarking sign recognition, defect presence, and delamination severity. Across the evaluated models, the findings indicate that defect-label quality and representative training examples impose stronger constraints than architectural differences. [Project Paper] [Conference proceeding]
A vision-based approach to assess the condition of highway traffic signs
Funded by the Utah Department of Transportation
Graduate Student: Shailaja Ratna Pandey
This project investigates how computer vision models can translate roadway imagery into useful information for transportation asset inspection and maintenance. In collaboration with the Utah Department of Transportation, we developed a condition-assessment framework that pairs MUTCD-aligned defect annotations with convolutional (YOLOv11) and transformer-based (RT-DETR) object detectors, benchmarking sign recognition, defect presence, and delamination severity. Across the evaluated models, the findings indicate that defect-label quality and representative training examples impose stronger constraints than architectural differences. [Project report] [Conference proceeding]
