Algorithms for Pavement Distress Classification by Video Image Analysis

A fundamental component of computer-based video image analysis for pavement distress evaluation is identification of the type of distress from geometric and textural properties of an area of interest identified during image analysis. This study describes the algorithms used to identi_fy and classify the most common pavement distress types once a possible distress region is identified on an image. The classification is accomplished in three steps. First, geometric and textural features are calculated for a region of interest. Next, the features of that region are used to determine whether any other regions in an image are actually part of the same pavement distress. Finally, the extracted image features are used with a decision tree to identify the specific pavement distress type, its severity, and its extent. The features and decision trees have been tested on several thousand pavement images and the system that contains the decision tree has been used successfully by the Ohio Department of Transportation since May 1994.