Steel Fatigue Crack Inspection of Civil Infrastructure Through Image Overlapping

KU researchers have developed a computer-based image overlapping approach for detecting structural defects in metallic structures.

Overview:

Computer vision-based crack detection methods have shown great potential as an alternative to visual inspection or use of fixed sensors.  However, many of these require advanced knowledge of cracks or struggle to distinguish between cracks and other features.  The KU computer vision-based crack detection method is a low-cost, accurate, and flexible technique that involves overlapping and aligning two images of a target structure subjected to repetitive loading to detect cracks and related features.

Applications:

The innovative crack detection technique detects structural defects of in-service, metallic physical structures such as bridges and roadways.  Early knowledge of such defects helps prevent damage due to corrosion, fatigue, and unexpected loading conditions.

How it works:

The KU innovation involves obtaining two images of the monitored structure from a consumer-grade digital camera.  Through a variety of image processing techniques, the difference in feature characteristics between the two images is extracted, enhanced, and visualized to identify any growing cracks and related features.

Benefits:

The method is low-cost to implement, more robust than edge detection techniques which often produce false positives, and flexible in its ability to account for varied camera angles (ideal for assessment of in-service structures and makes UAV platforms feasible).

Why it is better:

The KU detection method is compatible with consumer-grade digital cameras, eliminating requirements for special lighting, surface treatments, and sensor maintenance costs.  The method does not rely on detecting edge features like many approaches, which can have difficulty distinguishing true cracks from other features.  Testing showed that the approach can reliably identify fatigue cracks in the presence of non-crack features.  The method does not require advanced knowledge of crack details to train an algorithm.

Additional Details