Models for mapping from an Automatic Pavement Condition Survey to a Legacy Manual Survey

摘要

In the past decades, automatic pavement condition surveys, such as the Laser Crack Measurement System (LCMS), have matured reliably for wide adoption by transportation agencies. The LCMS often features a combination of high-resolution images and laser surface profiles, and provides reliable ratings of pavement conditions with saved time and cost. On the other hand, many agencies have maintained a comprehensive historical manual Visual Evaluation System (VES) that they are reluctant to completely abandon. In this paper, two models for creating a mapping rule from an automatic LCMS measure to its corresponding VES measure have been studied. First, a systematic process to transform a LCMS pavement condition dataset is proposed so that length-resolution of the transformed LCMS is compatible with that of the VES condition dataset. Second, through the use of statistical analysis and online PhotoLog, erroneous and outlier samples from the transformed LCMS dataset are identified and removed. Third, the ordinal logistic regression (OLR) model with step-wise variables selection is employed and the decision tree classification model is used to develop the mapping rule from the LCMS to the VES systems. Both the logistics regression and decision tree classification models are evaluated using confusion matrix in terms of misclassification, as well as using error tolerance range suggested by the transportation engineers.