P. Luescher, D. Burghardt, R. Weibel

University of Zurich, Department of Geography, Zurich, Switzerland


Multiple representation databases (MRDB) contain several geographic datasets covering the same area. In an MRDB, the objects within the different datasets that model the same real-world phenomenon (also termed homologous objects) are linked with each other. Thus MRDB allow better consistency control between integrated datasets and cheaper updating by propagating updates of one dataset automatically into all other datasets. Equally important is the potential for new visualisation methods, for example by allowing a seamless zoom in real-time between two interlinked scale levels.

There exists already a large quantity of datasets and these have to be integrated into MRDB as well. Since manual integration would often be too tedious and time-consuming, automatic matching techniques that find homologous objects and create the links between them have to be developed. While existent matching processes have been designed for datasets that are of the same or at least similar scales, the goal of our work was the development of such a matching technique for road data of greatly different scales, e.g. 1:25.000 and 1:200.000.

Several challenges arise. Firstly, the conceptualisation of real-world objects varies in different scale levels. For instance, roundabouts collapse to single points in smaller scales; highways are just portrayed by their centreline, while in larger scales usually single lanes are modelled. Secondly the larger scale road network is much denser, and thus many short road segments of the larger scale have to be assigned to one single road of the smaller scale. Thirdly, because smaller scale roads are strongly generalised, geometry of homologous objects may differ considerably.

The matching method developed by us works in three stages: An initial filtering of smaller scale features generates probable matching pairs; then links between crossroads are created; finally the crossroads assignments are converted into road assignments. The filter step could be enhanced significantly by employment of an algorithm termed closest path, a shortest path algorithm that minimizes the Hausdorff distance between a reference road of the smaller scale and the set of possible matching candidates of the larger scale. In the matching stage, match precipitation by means of a line tracing module is used.

The matching algorithm has been evaluated with road data at the scales of 1:25.000 and 1:200.000, respectively. A comparison of the results to manually matched reference data showed that the method worked very well in areas with low to medium density of settlements, but had some deficiencies in complex urban zones. Therefore we propose to add an initial classification into rural and urban areas, and to adapt the algorithm parameters accordingly.