Analysis of Error in Pavement Ground Truth Indicators for Evaluating the Accuracy of Automated Image Collection and Analysis System

Abstract

Abstract The main purpose of this paper is to present the Automated Image Collection System (AICS) and Manual Image Analysis System (MIAS) as “ground-truth” for evaluating Automated Image Collection System (AICS) and Automated Image Analysis System (AIAS) for highway asset management. The proposed “ground-truth” is evaluated with respect to its repeatability against a traditional “ground-truth” procedure based on Manual Distress Collection System (MDCS) and Manual Distress Analysis System (MDAS). To capture digital images automatically, a vehicle mounted with a digital video camera was driven on the pavement test section. To determine the repeatability of MIAS, the images were evaluated by three individuals twice per individual. To determine the repeatability of MDCS/MDAS, the same individuals were asked to evaluate the same pavement section in the field twice per individual. Repeatability on three crack types was evaluated for two survey methods and three individuals. Overall, the average relative precisions of AICS/MIAS-based procedure by three individuals were 10, 8, and 19 % for longitudinal, transverse, and block cracks, respectively, whereas those of MDCS/MDAS-based procedure were 43, 45, and 41 % for longitudinal, transverse, and block cracks, respectively. In conclusion, the proposed AICS/MIAS-based “ground-truth” measurements can be considered as more repeatable by human operators than MDCS/MDAS-based “ground-truth” measurements.