Automatic crack detection on road imagery using anisotropic diffusion and region linkage

A novel strategy to automatically detect cracks in road pavement surface imagery, acquired by a laser imaging system, is proposed. It includes a new procedure to link fragments of crack regions based on a maximum a posteriori (map) classifier, which relies on a set of geometric characteristics of the segmented binary regions. The proposed system starts with an anisotropic diffusion filtering, to smooth image texture variations resulting from the type of sensor used. Then, a Gaussian function is used to model the histogram for pixels intensities below a certain value, which allows inferring the threshold level to be used for image segmentation. Next, less relevant binary regions are identified, being kept only if they are linked to relevant crack regions. Encouraging experimental crack detection results are presented based on real images captured along Canadian roads.