Simultaneous Chain-Forming and Generalization of Road Networks

Streets are essential entities of urban terrain and their automatic extraction from airborne sensor data is cumbersome because of a complex interplay of geometric, topological, and semantic aspects. Given a binary image representing the road class, centerlines of road segments are extracted by means of skeletonization. The focus of this paper lies in a well-reasoned representation of these segments by means of geometric primitives, such as straight line segments as well as circle and ellipse arcs. Thereby, we aim at a fusion of raw segments to longer chains which better match to the intuitive perception of what a street is. We propose a two-step approach for simultaneous chain-forming and generalization. First, we obtain an over-segmentation of the raw polylines. Then, a model selection approach is applied to decide whether two neighboring segments should be fused to a new geometric entity. For this purpose, we propose an iterative greedy optimization procedure in order to find a strong minimum of a cost function based on a Bayesian information criterion. Starting at the given initial raw segments, we thus can obtain a set of chains describing long alleys and important roundabouts. Within the procedure, topological attributes, such as junctions and neighborhood structures, are consistently updated, in a way that for the greedy optimization procedure, accuracy, model complexity, and topology are considered simultaneously. The results on two challenging datasets indicate the benefits of the proposed procedure and provide ideas for future work.