A Unified Preprocessing Pipeline for Noise-Resilient Crack Segmentation in Leaky Infrastructure Surfaces

Area of Science:

• Civil Engineering
• Materials Science
• Computer Vision

Background:

• Wet cracks present visual distortions from contamination and corrosion, degrading sensor-based detection.
• Nonlinear crack propagation in moist conditions complicates distinction from background noise like stains and low contrast.

Purpose of the Study:

• To propose a robust segmentation framework with a dedicated preprocessing pipeline for enhanced crack detection in adverse conditions.
• To improve the accuracy and reliability of vision-based crack detection systems in real-world infrastructure inspections.

Main Methods:

• A preprocessing pipeline incorporating adaptive thresholding, morphological operations, and connected component analysis.
• Contrast enhancement techniques including histogram stretching and contrast limited adaptive histogram equalization.
• A background fusion step to emphasize crack features while preserving surface texture.

Main Results:

• The proposed method significantly enhances segmentation performance under challenging conditions.
• Achieved a precision of 97.5% with strong robustness against moisture, reflections, and surface irregularities.
• Demonstrated substantial improvement in accuracy and reliability for infrastructure inspection.

Conclusions:

• Targeted preprocessing is crucial for overcoming limitations in current crack detection systems.
• The developed framework offers a reliable solution for detecting cracks in visually complex and moist environments.
• This approach can substantially enhance the accuracy and reliability of crack detection systems in real-world infrastructure inspection scenarios.