2018 Volume 6 Issue 1 Pages 78-90
Human inspection of defects of concrete tunnel lining is usually slow, laborious, and disruptive to traffic. This necessitates automated alternatives using sensors and computer-aided processing. The conventional image-matching methods only use the cost value of the pixel being processed based on similarity metric to estimate an image-matching location. To improve the image-matching efficiency, this paper proposes an image-matching method that relies on the curvatures of the cost curve at candidate matching points. This is followed by applying a median filter to mitigate the matching errors. Moreover, experimental results for an actual tunnel demonstrate that the curvature measurement can select the matching points accurately. The authors have developed an image-stitching software to create a high-resolution panorama of the tunnel lining surface for assisting in defect inspection.