Automatic pavement distress detection system

Abstract

Statistics published by the Federal Highway Administration indicates that maintenance and rehabilitation of highway pavements in the United States requires over $17 billion a year. Conventional visual and manual pavement distress analysis approaches that the inspectors traverse the roads, stop and measure the distress objects when they are found, are very costly, time-consuming, dangerous, labor-intensive, tedious, subjective, having high degree of variability, unable to provide meaningful quantitative information, and almost always leading to inconsistencies in distress detail over space and across evaluations. This paper introduces a new pavement distress image enhancement algorithm, and a new analysis and classification algorithm. The enhancement algorithm corrects nonuniform background illumination by calculating multiplication factors that eliminate the background lighting variations. The new pavement distress classification algorithm builds a data structure storing the geometry of the skeleton obtained from the thresholded image. This data structure is pruned, simplified, and aligned, yielding a set of features for distress classification: number of distress objects, number of branch intersections, number of loops, relative sizes of branches in each direction, etc. This skeleton analysis algorithm relies on two-dimensional geometrical parameters, which are understandable by both developers and users, unlike some methods that deal with abstract quantities not readily understood by ordinary users. The proposed analysis algorithm can precisely quantify geometrical and topological parameters, can quickly accept new classification rules for classification, and can estimate the distress severity from the thresholded image. The experimental results are satisfactory.