Image-based crack detection and properties retrieval for high-speed railway bridge

Recently, image processing technology has been developed into the inspection of bridges to detect cracks. However, there are still some difficult problems unresolved, for example, noises in the concrete bridge pictures are very similar to cracks, may be even more dominant than cracks. In this paper, an improved method to handle this situation is proposed. A modified percolation model is used to detect cracks automatically, generating a crack map for automatic measurement. Then, a distance transform is employed to get the information of distance and store it into black pixels. Finally, a thinning algorithm is applied to the crack map to get the skeletons of cracks and extract data stored in pixels. The performance of the proposed method is evaluated by receiver operating characteristics analysis. Experimental results show that the proposed method can simulate the real data correctly.