Detecting healthy concrete surfaces.

Teams of engineers visually inspect more than half a million bridges per year in the US and EU. There is clear evidence to suggest that they are not able to meet all bridge inspection guideline requirements due to a combination of the level of detail expected, the limited time available and the large area of bridge surfaces to be inspected. Methods have been proposed to address this problem through damage detection in visual data, yet the inspection load remains high. This paper proposes a method to tackle this problem by detecting (and disregarding) healthy concrete areas that comprise over 80–90% of the total area. The originality of this work lies in the method’s slicing and merging to enable the sequential processing of high resolution bridge surface textures with a state of the art classifier to distinguish between healthy and potentially unhealthy surface texture. Morphological operators are then used to generate an outline mask to highlight the classification results in the surface texture. The training and validation set consists of 1028 images taken from multiple Department of Transportation bridge inspection databases and data collection from ten highway bridges around Cambridge. The presented method achieves a search space reduction for an inspector of 90.1% with a risk of missing a defect patch of 8.2%. This work is of great significance for bridge inspectors as they are now able to spend more time on assessing potentially unhealthy surface regions instead of searching for these needles in a mainly healthy concrete surface haystack. [ABSTRACT FROM AUTHOR]