What is AI-Based Crack Detection?
AI-based crack detection is a computer vision method that finds cracks in pavement images with a model that was trained on labelled examples.
How it works
People mark the crack pixels in a set of training images. The result is a mask for each image. A model learns from these pairs. Then it marks the crack pixels in new images. The CrackForest dataset is one public example. It has 118 pavement images with labelled masks.
Earlier methods used a brightness threshold. A threshold works on a clear background, but it labels many pixels wrongly when the background is noisy.
What a result needs for airfield work
Airfield pavement is rated by type, severity and quantity of each distress. For a longitudinal or transverse crack in asphalt, the PAVER manual sets the severity by spalling and by crack width. A non-filled crack with a mean width of 1/4 inch (6 mm) or less, and little or no spalling, is low severity. The crack is measured in linear feet or metres. So a result that only says “crack here” is not enough for a PCI survey. It must also give length and severity.
Cracks in flexible pavement have different causes. FAA AC 150/5380-6C lists deflection over an unstable foundation, shrinkage, thermal movement, poor lane joints and reflection cracking. It names longitudinal, transverse, block, reflection and alligator (fatigue) cracks.
Limits
- A detected crack can be wider than the real one. Research on UAV images reports that methods label the pixels next to a crack as crack.
- Scattered noise can remain after the threshold step.
- The FAA states that visual inspection alone cannot show the structural condition of a pavement.
- The sources checked for this page do not name AI image analysis as an accepted method for a PCI survey or an aerodrome inspection.
Where it fits
EASA guidance on the inspection of paved areas asks inspectors to look for cracking and spalling. FAA AC 150/5380-7B asks for pavement condition data that is objective, systematic and repeatable. Neither document names image analysis as a way to meet these needs.
Frequently Asked Questions
- How does AI-based crack detection work?
- A model learns from images in which people have marked the crack pixels. After training, it marks the crack pixels in new images. Older methods used a brightness threshold, and they labelled many pixels wrongly when the background was noisy.
- Is AI-based crack detection the same as a PCI survey?
- No. A pavement condition index (PCI) survey follows ASTM D5340 and records each distress by type, severity and quantity. A crack detector only finds cracks in an image. A PCI survey needs crack length and severity as well.
- Do aerodrome rules accept AI crack detection?
- The sources checked for this page do not mention it. EASA guidance asks for a check for cracking during inspection of paved areas. The FAA pavement management circular asks for an objective, systematic and repeatable evaluation. Neither document names an image-analysis method.
References
- zhu-2019-hcnn
- Crack Detection Using Enhanced Hierarchical Convolutional Neural Networks (arXiv:1912.12139) Zhu, Phung and Ha, University of Technology Sydney
- shi-2016-cfd
- Automatic Road Crack Detection Using Random Structured Forests, IEEE Transactions on Intelligent Transportation Systems 17(12), 2016 IEEE
- faa-paver-asphalt
- PAVER Asphalt Surfaced Airfield Pavements Distress Identification Manual FAA, US Army Corps of Engineers
- faa-ac-150-5380-6c
- AC 150/5380-6C, Guidelines and Procedures for Maintenance of Airport Pavements FAA
- faa-ac-150-5380-7b
- AC 150/5380-7B, Airport Pavement Management Program (PMP) FAA
- easa-ear-aerodromes
- Easy Access Rules for Aerodromes (Regulation (EU) No 139/2014), Part-ADR.OPS EASA
This is a reference entry. TarmacView provides drone inspection services and doesn't sell the equipment or materials described in this glossary.
