Project Details

Abstract

This project develops a scalable, AI-powered framework for automated condition assessment and capital investment prioritization of road and bridge infrastructure across Colorado. Current manual inspection methods are costly, slow, and limited in coverage, often missing early signs of degradation. Leveraging recent advances in Vision-Language Models (VLMs), the project proposes a novel approach that extracts Pavement Condition Index (PCI) and bridge deck ratings from satellite and street-level imagery using VLMs guided by prompt engineering and in-context learning, requiring no retraining, and will be validated against Colorado Department of Transportation (CDOT) inspection records using machine learning model accuracy metrics.

The framework integrates three components: (1) Infrastructure condition (2) network criticality, computed via graph-theoretic metrics and traffic data, and (3) hazard exposure, based on geospatial data for landslides, wildfires, and floods. These layers will be synthesized into a weighted prioritization model to rank road segments for capital upgrades, refined through CDOT expert review.

A web-based GIS tool will visualize prioritization results, supporting interactive exploration and decision-making. The tool will be pilot-tested with CDOT districts, and outcomes will be disseminated through a final report, peer-reviewed publication, Transportation Learning Network (TLN) webinar, and Colorado LTAP training.

By integrating AI, geospatial analytics, and network modeling, this project addresses data-driven infrastructure planning needs—enhancing efficiency, reliability, and resilience—while offering a replicable model nationwide.

Project Word Files