Automatic Crack Detection by Multi-Seeding Fusion on 1mm Resolution 3D Pavement Images

2D image-based crack detection has been studied for many years. But we are still short of a robust method, which can generate satisfied results on general cases. 3D pavement images have obvious advantages over 2D images in dealing with disturbances caused by lane markings, sun-light shadows, oil marks and debris. Taking advantages of the state-of-the-art 3D image acquisition technology, this paper proposes a novel crack detection algorithm. It has four steps: First, generate two partly overlapped images with 8×8 pixel grid cells. Second, extract 10 images of crack seeds through symmetry check and grayscale verification. Third, connect scattered seeds by an optimal path searching process considering grayscale values, cell direction and proximity. Finally, 10 detection results from two grid cell images are combined and denoised to obtain final results. Experiments based on 166 test images show that the proposed method can achieve better performance measured by F-measure, which takes both precision and recall rates into consideration.