Automatic Pixel-Level Pavement Crack Detection Using Information of Multi-Scale Neighborhoods

Robust automatic pavement crack detection is critical to automated road condition evaluation. However, research on crack detection is still limited and pixel-level automatic crack detection remains a challenging problem, due to heterogeneous pixel intensity, complex crack topology, poor illumination condition, and noisy texture background. In this paper, we propose a novel approach for automatically detecting pavement cracks at pixel level, leveraging on multi-scale neighborhood information, and pixel intensity. Using pixel intensity information, a probabilistic generative model (PGM) based method is developed to calculate the probability of a crack for each pixel. This produces a probability map consisting of the probability of each pixel being part of the crack. We demonstrate that the neighborhoods of each pixel contain critical information for crack detection, and propose a support vector machine (SVM) based method to calculate the probability maps using information of multi-scale neighborhoods. We develop a fusion algorithm to merge the multiple probability maps, obtained from both PGM and SVM approaches, into a fused map, which can detect cracks with accuracy higher than any of the original probability maps. We also propose a weighted dilation operation that relies on the fused probability map to enhance the recognition of borderline pixels and improve the crack continuity without increasing the crack width improperly. Experimental results demonstrate that our algorithm achieves better performance in terms of precision, recall, f1-score, and receiver operating characteristic, in comparison with the state-of-the-art pavement crack detection algorithms.