Author(s): Norman Wittels, Tahar El-Korchi, Michael A. Gennert, and Matthew O. Ward, Worcester Polytechnic Institute, Worcester, MA 01609

Source: Proc. Automated Pavement Distress Data Collection Equipment Seminar, pp153-164, Iowa State University, Ames IA, 12-15 June 1990.

Abstract: Automated pavement surface distress evaluation systems do not always perform as well during field use as they do in laboratory tests because the test pavement images do not represent the full range of signals that the system can see. The first task of the image processing computer in such a system is to determine whether each individual picture element, pixel, is the image of a crack or of sound pavement. The probability of correctly identifying a crack increases monotonically with the video signal contrast between images of crack and pavement; high contrast cracks are easier to detect. In previous work we have shown how the crack contrast depends on the paving materials, crack geometry, and lighting.

The normal variations in materials, geometry, and lighting encountered while inspecting pavements can cause crack image contrast to diminish or even reverse; the cracks are then brighter than the pavement surface. Therefore, when selecting the test pavement images for validating inspection system performance, it is important to make selections that test the system performance over the full range of image contrast.

In this paper we show how to document pavement images that are being acquired for test purposes, and how to select among or alter acquired images to optimize the testing of automated systems.