United States Department of Transportation
i
Mobile Phone-Based Artificial Intelligence Development for Maintenance Asset Management
-
2022-10-01
[PDF-2.17 MB]
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report July 2021 to October 2022
-
Corporate Publisher:
-
Abstract:Transportation asset management needs timely information collection to inform relevant maintenance practices (e.g., resource planning). Traditional data collection methods in transportation asset management require either manual operation or support of unique equipment (e.g., Light Detection and Ranging (LiDAR)), which could be labor-intensive or costly to implement. With the advancement of computing techniques, artificial intelligence (AI) has been developed to be capable of automatically detecting objects in images and videos. In this project, we developed accurate and efficient AI algorithms to automatically collect and analyze transportation asset status, including identification of pavement marking issues, traffic signs, litter & trash, and steel guardrails & concrete barriers. The AI algorithms were developed based on the You Only Look Once (YOLO) framework built on Convolution Neural Network as the deep learning algorithms. Specifically, a smartphone was mounted on the vehicle’s front windshield to collect videos of transportation assets on both highways and local roads. These videos were then converted and processed into labeled images to be training and test datasets for AI algorithm training. Then, AI models were developed for automatic object detection of the listed transportation assets above. The results demonstrate that the developed AI models achieve good performance in identifying targeted objects with over 85% accuracy. The developed AI package is expected to enable timely and efficient information collection of transportation assets, hence, improving road safety.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
