Automatic Extraction and Evaluation of Road Networks from Moms-2p Imagery

In this paper an approach for the automatic extraction and evaluation of road networks from MOMS-2P imagery is proposed. Due to the limited spatial resolution of the images for the specified task a road model purely based on local criteria is rather weak, and therefore a significant number of false alarms are to be expected. A model is defined based on the local, regional, and global properties of the road network and a corresponding extraction strategy is derived. The extraction strategy incorporates line extraction using a differential geometric approach, followed by constructing a weighted, planar graph from the lines and the gaps between them according to local (radiometric) and regional (geometric) criteria. By computing various “best paths” through this graph the actual network is derived based on global (topological) criteria. Line extraction is carried out for all available channels, and the individual results are fused prior to grouping. Thus, the information from multiple spectral channels is exploited. The evaluation of the results is carried out by comparison to manually derived reference networks showing the feasibility but also the problems of the approach.