United States Department of Transportation
ROSA P serves as an archival repository of USDOT-published products including scientific findings, journal articles, guidelines, recommendations, or other information authored or co-authored by USDOT or funded partners.
As a repository, ROSA P retains documents in their original published format to ensure public access to scientific information.
i
Mobile-Phone-Based Artificial Intelligence Package Development and Validation in Large Scale for Maintenance Asset Management
-
2024-10-01
-
[PDF-4.85 MB]
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report May 2023 to October 2024
-
Corporate Publisher:
-
Abstract:Regular inspection of transportation assets is essential to maintain their optimal condition and functionality. Our previous UDOT-funded project (Contract No. 22-8099, Report No. UT-22.24) demonstrated significant promise of artificial intelligence in identifying various transportation assets through a mobile phone. However, the initial project was limited in scale, utilizing only ~1,000 images for training and validation. Therefore, this new project aims to extend our prior work. Specifically, this study increases the dataset to ~5,000 images for each targeted asset (i.e., pavement markings, litter & trash, traffic signs, and guardrails & barriers). Based on You Only Look Once (YOLO), a deep learning architecture, we trained four detection models capable of automatically identifying these objects with good accuracy metrics: F1 scores of 0.88 for pavement marking issues, 0.84 for litter/trash, 0.91 for traffic signs, and 0.96 for guardrails/barriers. Additionally, specific counting and geolocation models were developed to precisely quantify the number of identified objects within a road section or video clip and pinpoint the location of each detected object with the assistance of a phone-based GPS tracker. The geolocation model exhibits high performance in estimating object locations, with an average distance error of only 0.27 meters (about 0.9 feet). Furthermore, an interactive interface was created to visually represent the identified objects on a map, allowing for a comprehensive assessment of transportation asset conditions through intuitive visualizations. This new project enhanced our previous work by expanding the capabilities in detecting, counting, geolocating, and visualizing specific transportation assets, contributing to implementing regular transportation asset inspection and planning maintenance work, and thereby improving road safety.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
