This paper addresses the problem of automatically generating city street networks from a possibly inaccurate vector description of their blocks. The street network, represented as a graph structure with attributes related to both streets and crossings, is a fundamental data structure for a large number of applications, including urban planning and transport management. The considered inaccuracies derive from the typical errors introduced during hand digitizing; i.e., undershooting, overshooting, stacking, and layer misplacing. Extracting the street net when all these possible errors are present is not an easy task because of the huge amount of particular cases to be considered. The presented algorithm is based on the translation of the vector information directly into a run-length encoded binary image. Then, corrections of the above errors and of the street net extraction can be expressed in terms of well known, simple, morphological operations. In order to lower computational costs, the key idea has been to carry out these operations directly on the encoded image. The result is an algorithm that is faster than expected, easy to implement, and, as a consequence, more generally applicable.

This work has been partially supported by ENHER (Empresa Nacional Hidroelectrica del Ribagorzana) under several contracts within the framework of the ALEPH-CORAL project.

Peer Reviewed