Anwendung automatischer Extraktionsverfahren von Straßenverkehrsnetzen in urbanen Räumen

Kurzfassung

From urban areas today exist many high-resolution aerial photographs and their potential for geoscientific studies in most cases cannot be fully exploited. An example is extraction of road surface on the basis of data obtained by remote sensing. For this case there are no such reliable extraction methods. That can be dispensed with manual corrections and additions in the detection of road networks. This study shows how an automated extraction can be carried out by road surfaces. One possibility to capture roadways in urbanized areas is the utilization of object-based classification. Remote sensing data are combined into pixel groups and assigned on the basis of homogeneity criteria classes. This method allows, in contrast to other classification methods such as pixel-based classification, on the basis of data at a comparatively low information content, to arrive at effective surface detection. Basis of this work were many researches for the automatic detection of all kinds of surfaces. In this method, which appeared from the culling of road space when applicable, tested and applied to a positive result. The data base is represented by two views of Vaihingen an der Enz (Baden-Wuerttemberg) and Guadalajara (Mexico). Each has a standardized digital surface model. Three- or fourchannel high-resolution aerial images were available. The aerial photographs were counted before beginning work on a true-ortho-mosaic. This improved the quality of the recordings and use of complete regions from the zenith view allowed. The extraction procedures were implemented and tested. These were specifically developed for object-based classification of raster data in the software Trimble eCognition. The Accuracy Assessment Tool of the software Erdas Imagine was used for the evaluation. The result of the extraction method employed was used as the basis for work. It was not as effective as the significant improvement results from older works had been. It could, however, be shown that the analysis of large urban areas in an acceptable period of time is certainly possible. The view of the city of Guadalajara included, for example, 24 km2 at a resolution of 15 cm, which corresponds to about one billion pixels. Finally some functions could be tested for their usefulness and identified, at that time, to be ineffective.