This thesis presents an approach for the automatic extraction of road networks from optical satellite imagery and for the evaluation of the extraction results. For the extraction of the roads, road network characteristics are explicitly used. The evaluation comprises the quality of the extracted roads, using measures, known from literature. The network characteristics themselves are evaluated based on newly developed quality measures. Tests show that the use of the network characteristics for road extraction leads to results with much better network characteristics. Input data are optical aerial or satellite images with a ground pixel size of about 2 m x 2 m, possibly having more than one spectral channel. The approach for road extraction consists of two parts: the preliminary extraction and the improvement of the extraction. In the first part, a preliminary road network is extracted. The used model comprises local (radiometric) and regional (geometric) aspects, as well as the connection property of roads (topological aspects). This property of road networks is implemented through the search for long connected road segments. In the second part, the improvement of the extraction result based on further network characteristics is performed. On the one hand, among ancillary conditions, the road network is optimized such that there exists a connection as short as possible between any pair of points lying on the road network. This is a major reason for the network structure of roads and, therefore, an important network characteristic. This characteristic is implemented by locally analyzing the path lengths within the extracted road network. On the other hand, the network structure of roads depends on the crossings which connect roads. By means of an explicit reconstruction of crossings, some errors of the preliminary extraction are eliminated and, a visually pleasing and topologically correct result can be obtained. The evaluation of the extraction results is carried out by a comparison with reference data. First, the extraction results are matched to the reference data. Then, quality measures for the evaluation of the extraction are derived from the matching results. The evaluation is carried out separately for roads and crossings. A new approach for the evaluation of the network characteristics of the extraction results is presented in this thesis. It is based on the analysis of path lengths as well as on the topology of the extraction results, in each case with respect to the reference data. It supplements the existing approaches for the evaluation of extraction results by adding the evaluation of network characteristics which are especially important for road networks. The complete process is illustrated on the basis of one example and the results are evaluated and discussed. Subsequent, the strengths and weaknesses of the developed approaches are discussed. The obtained results prove the capability of the developed approaches for the extraction of road networks. An outlook concludes the thesis.
«
This thesis presents an approach for the automatic extraction of road networks from optical satellite imagery and for the evaluation of the extraction results. For the extraction of the roads, road network characteristics are explicitly used. The evaluation comprises the quality of the extracted roads, using measures, known from literature. The network characteristics themselves are evaluated based on newly developed quality measures. Tests show that the use of the network characteristics for ro...
»