Fully Convolutional Networks for Automatic Pavement Crack Segmentation

Resumen

Pavement cracks are an increasing threat topublic safety. Automatic pavement crack segmentationremains a very challenging problem due to cracktexture inhomogeneity, high outlier potential, largevariability of topologies, and so on. Due to this,automatic pavement crack detection has captured theattention of the computer vision community, and agreat quantity of algorithms for solving this task havebeen proposed. In this work, we study a U-Netnetwork and two variants for automatic pavement crackdetection. The main contributions of this research are:1) two U-Net based network variations for automaticpavement crack detection, 2) a series of experiments todemonstrate that the proposed architectures outperformthe state-of-the-art for automatic pavement crackdetection using two public and well-known challengingdatasets: CFD and AigleRN and 3) the code for thisapproach.