Statistical Approach to Road Segmentation

Abstract

A method for segmentation of a road based on surface measurements is presented. This method is based on a statistical model of the measurement series and acknowledges that neighboring measurements are dependent. Sudden changes in the level, in the variance, or in the autocorrelation of a series are detected. No prior knowledge about these quantities is required, and no distributional assumptions about the nature of the sudden change are made. The method allows for an assessment of the information contained in the measurements themselves. Uncertainties about the existence and possible location of a change-point are communicated to the user in terms of probabilities. The road engineer may then match the extracted information with available complementary information about, e.g., the age of the pavement and traffic volumes. The focus is on the situation where it is known beforehand that, at most, one change points is present. Thus, the suggested method is best suited to assist the engineer in a detailed study of selected parts of a given measurement series, or may be used iteratively when several change points are suspected. The application of this method is demonstrated using a measurement series for the international roughness index, collected with the Swedish road surface tester known as the laser RST-vehicle.