Automated Road Segmentation Using a Bayesian Algorithm

Abstract

Modern road profilers deliver long sequences of measurements on road characteristics including a road’s longitudinal and transversal unevenness. These measurements represent adjacent parts of the physical road, and interest focuses more on the overall pattern of these measurements than on each single value. In order to systematically assess the information contained in these measurement series, one typically wishes to partition a given series into segments, where each segment contains measurements which are “similar” to each other but “dissimilar” to the elements in the neighboring segments. An algorithm is suggested that combines a recently developed Bayesian identification of transitions between two homogeneous road sections with a heuristic approach that uses this technique iteratively to find multiple homogeneous sections in arbitrary long measurement series. The approach is demonstrated with narrowly spaced measurement series of the international roughness index as well as rutting.