Quantification of Structural Defects Using Pixel Level Spatial Information from Photogrammetry.

Area of Science:

• Civil Engineering
• Structural Health Monitoring
• Computer Vision

Background:

• Aging infrastructure poses significant economic and social risks globally.
• Accurate quantification of minor defects is crucial for preventing structural failure.
• Existing methods often overlook 3D information from photogrammetry, limiting defect measurement accuracy.

Purpose of the Study:

• To develop an efficient and accurate method for detecting and quantifying minor defects on complex infrastructure.
• To integrate 3D spatial information with deep learning for precise defect dimension estimation.
• To overcome limitations of traditional image-level defect analysis.

Main Methods:

• Estimating pixel size using 3D point cloud reconstruction for spatial information.
• Applying deep learning (Convolutional Neural Network - CNN) for pixel-level crack detection.
• Calculating actual crack dimensions by multiplying pixel count with estimated pixel size.

Main Results:

• The CNN achieved an F1 score of 0.613 for minor crack extraction.
• A pilot study on a concrete footpath demonstrated average errors from 0.26 mm to 0.71 mm for cracks under 5 mm.
• The proposed approach accurately estimated defects on a complex structure.

Conclusions:

• The developed method enables precise quantification of minor defects on infrastructure by incorporating 3D spatial data.
• This approach offers a significant improvement over traditional methods for defect assessment.
• The findings show promising results for enhancing structural health monitoring and maintenance strategies.