Background
Culverts are buried and often hard to access, making traditional visual inspection challenging. To address these challenges, the use of digital video inspections by deploying cameras has emerged as a valuable tool in assessing culvert conditions. However, reviewing and interpreting the digital video from culvert inspections can be labor-intensive and requires highly trained personnel. Assessments rely on human judgment, leading to inconsistent condition ratings and potential errors.
A previous UDOT research project showed that an automated culvert inspection interpretation system using AI and computer vision can reduce the time, labor, and subjectivity involved in manually reviewing inspection videos, while supporting more consistent condition assessment. This can lead to better management of culverts as transportation assets.
Research
This WTRC research, building on work completed for a previous Utah DOT study, will use an expanded dataset drawing from additional culvert inspection videos from other western states. This will support the development of a more generalizable base model that individual states can customize to fit their own culvert inspection manuals and rating standards.