Crack Junction Detection in Pavement Image Using Correlation Structure Analysis and Iterative Tensor Voting

Crack junction is the crossing or branching point of different cracks in the pavement image. It represents the branch of transverse crack or longitudinal crack, and describes the interlaced network of alligator crack. It is a simple yet important factor to characterize the type and severity level of crack. This paper is motivated to robustly detect crack junctions of any type and size in pavement image, regardless of the pavement interferences. In this paper, the contrast between the crack junction and pavement background is first enhanced by removing the large interferences and background. Then, based on the structure characteristic of crack curves, correlation structure index is proposed to locate candidates of crack junctions. Actual junctions are extracted among the candidates with the unified ball tensor structure after the iterative tensor voting. The proposed method is tested with the concrete pavement images of public data set of SDNET2018 and asphalt pavement images collected by the unmanned aerial vehicle on the highway G45 in China. Experimental results demonstrate that the proposed method can detect crack junctions with the correctness of 0.891 and completeness of 0.887. It can be applied to junction detection on concrete and alligator pavement with different noise and interference, and is promising to classify the crack type and quantify the severity level.