Study On Bridge Crack Detection Algorithm Based On Image Processing Technology

Crack is a sign of damaged concrete bridge image, crack detection plays animportant role in bridge mass detection. At present, the artificial method has beenunable to meet the growing needs of bridge crack detection method mainly, thedisadvantage of which is high cost, high risk, and the efficiency of which is very lowas well. In recent years, bridge crack detection method based on digital imagesegmentation technology is playing an increasingly important role, among them,watershed segmentation algorithm has become a hot topic in the field of imagesegmentation because it is sensitive to the edge information, but it is also too sensitiveto noise information, prone to the "over-segmentation" phenomenon, so the keyquestion of this study is to bridge cracks image for the study, watershed segmentationalgorithm has been improved in the pre-processing and segmentation stages, therebyinhibiting the over-segmentation problems arising.Bridges image preprocessing: due to the cracks may produce in bridge pavement,pillar and the bottom of the other positions, the bridge crack images often have variousnoises, pits and contrast of which is also not strong, which makes it difficult to getinformation through direct segmentation method of crack. In the preprocessing stage,we propose a kind of morphology filter which has the multi-angle and multi-structureelement, which can remove image noise bridge cracks information effectively, andkeep the fine structure of the edge image effectively.Bridges image segmentation: in order to suppress the over-segmentation problemmore thoroughly, we present a marker watershed segmentation algorithm based on thetraditional watershed segmentation algorithm, the algorithm can remove a large part ofthe local minimum of the image, the segmentation results were greatly improved, but itcan not automatically marking when marking number, and the border of the accuracywill be reduced, and the over-segmentation phenomenon still exists; in order to restrain the over-segmentation phenomenon more thoroughly, this paper proposed anadaptive labeling watershed algorithm, the segmentation results were further improved,but there is still a small part of the noise information was split out; finally, this thesisproposes an adaptive multi-scale watershed segmentation algorithm, multi-scalegradient transformation is carried out for filtered bridge crack image firstly, thenwatershed transformation based on adaptive marker is used for gradient image, lastly,the final image is obtained through improved region merging, so that the over-segmentation phenomenon can be suppressed effectively. At last, the bridge crackimage segmentation algorithm was simulated on MATLAB.