A crack detection and quantification method using matched filter and photograph reconstruction.

Area of Science:

• Civil Engineering
• Computer Vision
• Structural Health Monitoring

Background:

• Bridge maintenance relies on accurate crack detection, but deep learning methods require extensive datasets.
• Traditional digital image processing offers interpretability and lower computational costs, warranting further research.

Purpose of the Study:

• To develop an automatic crack detection and quantification method for bridges using digital image processing and UAV parameters.
• To create an integrated 3D model of bridge damage for improved maintenance planning.

Main Methods:

• Analysis of UAV-collected bridge images.
• Enhanced matched-filter algorithm for crack segmentation.
• Morphological methods for crack skeleton extraction and length calculation.
• 3D model construction using detection results and UAV flight parameters.

Main Results:

• The enhanced matched-filter algorithm achieved 97.9% Pixel Accuracy, 72.5% F1-score, and 58.1% Intersection over Union on public datasets.
• The methodology was successfully applied to an arch bridge with only 2% error.
• The 3D model provided an intuitive representation of bridge damage.

Conclusions:

• The proposed digital image processing approach effectively detects and quantifies bridge cracks.
• Integration with UAV parameters enables comprehensive 3D damage modeling for informed decision-making.
• This method offers a practical alternative to deep learning for bridge structural health monitoring.