Crack detection on asphaltic concrete road surface images using modified grid cell analysis

Abstract

This research presents an image processing algorithm based on grid cell analysis for crack detection on asphaltic concrete road surface images. The research focuses on using pavement images from an area scan camera where light condition tend to be non-uniform and handling problems of strong texture of asphaltic concrete surface. The algorithm uses enhanced grid cell analysis to extract crack lines from images. Then, gravitational force feature is applied to remove noise. Finally, crack objects are merged using convex hull technique to find crack characteristics which are later used to classify crack types. The accuracy of the proposed method was measured by testing the algorithm with pavement images according to the research scope. The accuracy in finding cracked lines on pavement images with non-uniform illumination and strong texture could be achieved with an average value of 3.07% false positive and 9.17% false negative. Moreover, the test of the proposed method in real situation was done on pavement images from the open environment survey. The dataset was manually screened so that wet pavement images and underexposed images were excluded. The dataset was then processed, resulting in 6.2% false positive and 14.86% false negative respectively.