Automated Detection of Multiple Pavement Defects

Knowing the pavement condition is essential for efficiently deciding on maintenance programs. Current practice is predominantly manual with only 0.4% of inspections happening automatically. All methods in the literature aiming at automating condition assessment focus on two defects at most, or are too expensive for practical application. In this paper, the authors propose a low-cost method that automatically detects pavement defects simultaneously using parking camera video data. The types of defects addressed in this paper are two types of cracks (longitudinal and transverse), patches, and potholes. The method uses the semantic texton forests (STFs) algorithm as a supervised classifier on a calibrated region of interest (myROI), which is the area of the video frame depicting only the usable part of the pavement lane. It is validated using data collected from the local streets of Cambridge, U.K. Based on the results of multiple experiments, the overall accuracy of the method is above 82%, with a precision of more than 91% for longitudinal cracks, more than 81% for transverse cracks, more than 88% for patches, and more than 76% for potholes. The duration for training and classifying spans from 25 to 150 min, depending on the number of video frames used for each experiment. The contribution of this paper is dual: (1) an automated method for detecting several pavement defects at the same time, and (2) a method for calculating the region of interest within a video frame considering pavement manual guidelines.This material is based in part upon work supported by the National Science Foundation under Grant Number 1031329.This is the author accepted manuscript. The final version is available from the American Society of Civil Engineers via https://doi.org/10.1061/(ASCE)CP.1943-5487.000062