Over the years, many Automated Image Analysis Systems (AIAS) have been developed to objectively measure the amounts and severities of various pavement distress types. The purpose of this paper is to present a lane marking detection algorithm adopted for the AIAS called "PicCrack". The basic premise for the lane marking detection algorithm is that the lane markings, either yellow or white, would be brighter than the surrounding pavement surface. Therefore, to remove the lane markings from the pavement image, it was necessary to determine the relative average brightness of lane markings in comparison with the surrounding pavement surface. The lane marking detection threshold value was determined by multiplying the average gray-scale of the pavement surface by a parameter. The lane marking detection algorithm based on the multiplication parameter improved the accuracy of PicCrack. Further, the developed lane detection algorithm was validated against the randomly selected images which were collected under varying lighting conditions. Based on the test result of these images, the average value of the absolute differences between the true modified UCI and the modified UCI values using PicCrack was decreased when the lane marking detection algorithm was used. This result indicates that the lane detection algorithm has significantly improved the accuracy of PicCrack. The absolute difference between the true modified UCI and PicCrack with the lane detection algorithm is very close to the error caused by the background noises only. This result indicates that PicCrack with the lane marking detection algorithm is nearly as accurate as the manual analysis of the images with background noises.
Abstract