Abstract

This project aims to develop an interactive, web-based system to assist bridge inspectors in machine learning and video analytics for a rapid condition state assessment of bridge elements in accordance with the 2019 Manual for Bridge Element Inspection. Referred to BridgeNet, the system includes region-based convolutional neural network (RCNN) models for bridge element and defect segmentation and an inspection video analytic tool that is integrated into a unified graphical user interface (GUI). It provides users with an image-based testbed of modulated RCNN models. The BridgeNet has evolved in four phases: (1) backend and frontend framework setups, (2) Mask-RCNN-based bridge segmentation, (3) containerization of the system using Docker, and (4) multitask learning for element and defect segmentation. Docker was used to containerize the system and ensure its portability and scalability. Doing so made the system easier to deploy across various platforms. The multitask model showed promising results in identifying structural elements and defects such as corrosion with high accuracy. The learning module is also implemented to help users to understand the tools. Future improvements include enhancing crack detection capabilities and further integrating image processing algorithms. This system can significantly improve the precision, scalability, and automation of bridge inspections.

Department(s)

Civil, Architectural and Environmental Engineering

Second Department

Engineering Management and Systems Engineering

Research Center/Lab(s)

INSPIRE - University Transportation Center

Sponsor(s)

Office of the Assistant Secretary for Research and Technology U.S. Department of Transportation 1200 New Jersey Avenue, SE Washington, DC 20590

Comments

Principal Investigator: Genda Chen, Ph. D., P. E.

Co-Principal Investigator: Ruwen Qin, Ph. D.

Grant #: USDOT # 69A3551747126

Grant Period: 11/30/2016 - 09/30/2024

Project Period: 03/15/2021 - 09/30/2024

This investigation was conducted under the auspices of the INSPIRE University Transportation Center.

Keywords and Phrases

Neural network, multitask learning, video analytics, element segmentation, defect classification

Report Number

INSPIRE-017

Document Type

Technical Report

Document Version

Final Version

File Type

text

Language(s)

English

Rights

© 2025 Missouri University of Science and Technology, All rights reserved.

Publication Date

September 30, 2024

Download

Share

 
COinS