Abstract: The crack extraction by UAV images is one of the hot issues in recent years,for the problem of strong edge information interference of UAV image, a crack recognition method based on GrabCut operator is proposed. Firstly, the method uses GrabCut operator to extract the foreground pavement with cracks, and then uses denoising, edge detection and double threshold contour recognition methods to detect pavement cracks. This method eliminates a lot of false edge information and secondary noise interference, and realizes automatic crack recognition of high resolution UAV image. The experimental results demonstrate that the road surface extraction method based on GrabCut operator is superior to color extraction algorithm and watershed algorithm, and it is suitable for road surface extraction in complex scene, it is with high universality. At the same time, the method proposed in this study can quickly obtain fracture information, the detection scale can be manually controlled, and it is easy to realize multi-scale fracture information recognition. The research results can be applied to pavement crack location and identification, linear pavement facility detection, pavement disaster assessment and other fields.

Key words: GrabCut operator, watershed algorithm, Gaussian mixture model, computer vision, canny edge detection