Analysis of Automatic Road Extraction Results from Airborne Sar Imagery

Automatic extraction of roads is a present research topic. Many applications like topographic mapping, navigation applications, or image registration could profit from such algorithms. This paper is concerned with automatic road extraction from synthetic aperture radar (SAR) imagery. An approach originally developed for the extraction of roads in rural areas from optical imagery with a ground pixel size of about 2 m is evaluated based on the comparison of the extraction results with reference data. Three large test sites were used, for which high-resolution SAR data (0.5 – 2 m), topographic map data, as well as manually plotted reference data are available. Both, the map data and the manually plotted reference data, are separated into three classes: highways, main roads, and secondary roads. The comparison shows that the extraction results strongly depend on the road classes: For main roads quite satisfying results can be achieved. Also, for highways the results are acceptable, with the restriction that no strong scattering objects, like traffic signs or bridges, interfere the road. The results for secondary roads from the 2 m E-SAR imagery are rather incomplete, due to the low visibility. In case of the highresolution AeS-1 SAR imagery interfering objects of the industrial scenery lead to an incomplete extraction of secondary roads. * Corresponding author