Back to search

University of British Columbia

Development and application of monocular 3D point cloud for structural damage quantification

Abstract

dc:description

The early detection of structural damage is critical for preventing catastrophic failures and ensuring public safety. Traditional manual inspection methods are labor-intensive, time-consuming, and pose inherent safety risks to human inspectors. Consequently, the automation of inspection processes using advanced computer vision and machine learning techniques offers a promising alternative that is both efficient and safer. Currently, most computer vision-based inspection systems rely on two-dimensional (2D) analysis. However, 2D methods are inherently limited, as they primarily enable qualitative damage classification, basic localization, and in-plane quantification. The introduction of three-dimensional (3D) vision-based techniques has significantly expanded the capabilities of structural assessment by incorporating depth perception, a wider field of view, and improved performance in complex environments. Technologies such as laser scanning, stereo vision, photogrammetry, and structure-from-motion have been employed to generate high-resolution 3D point clouds for bridges and other structures. While effective, these methods typically require specialized equipment and are often constrained by slow data acquisition and processing speeds. This study investigates state-of-the-art deep learning-based methods for 3D reconstruction of structural elements from a single RGB image. Initially, monocular depth estimation algorithms are employed to generate depth maps from individual RGB images. This data is then converted into RGBD format, which is subsequently used to generate 3D point clouds. Structural damage is then autonomously identified and quantified within these point clouds using mathematical algorithms. A key requirement for accurate 3D reconstruction is knowledge of the intrinsic parameters of the camera. Two approaches were proposed for this purpose. The first involves standard camera calibration techniques widely used in computer vision. The second leverages deep learning-based estimation or approximation of intrinsic parameters. Experimental results demonstrate that intrinsic parameters can be estimated with sufficient accuracy, enabling the use of uncalibrated, previously captured images for structural damage assessment. The proposed methods offer a cost-effective, rapid, and scalable solution for both real-time and retrospective damage detection. Continued advancements in deep learning models for depth estimation and camera parameter prediction are expected to further enhance the accuracy and applicability of these techniques within the field of structural health monitoring.

Degree

thesis:*
Name thesis:degree_name
Master of Applied Science - MASc
Level thesis:degree_level
master's
Discipline thesis:degree_discipline
Civil Engineering
Grantor dc:publisher
University of British Columbia
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Faraji Zonouz, Elmira
Contributors dc:contributor
  • Yang, Tony T. Y.

Rights

dc:rights
Statement dc:rights
  • Attribution-NonCommercial-NoDerivatives 4.0 International
Language dc:language
eng

Identifiers

dc:identifier.*
Handle dc:identifier
http://hdl.handle.net/2429/90877

Chain of custody

source
Harvested from
University of British Columbia
Base URL
circle.library.ubc.ca/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Faraji Zonouz, Elmira. Development and application of monocular 3D point cloud for structural damage quantification. master's thesis, University of British Columbia, 2025. http://hdl.handle.net/2429/90877