A Fusion Strategy for Extracted Road Networks from Multi-aspect Sar Images

In this paper, we describe an extension of the automatic road extraction procedure developed for single SAR images towards multiaspect SAR images. Multi-aspect images illuminate the same scene, but from different directions. For the combination of the extracted information, a fusion technique is introduced. Each road segment is assessed according to its direction compared to the direction of the illumination. Due to the side-looking geometry of SAR, the visibility of roads is often limited by adjacent trees or building rows. Roads in viewing direction are less affected by shadow and layover effects from neighbouring objects than roads across the viewing direction. Road segments are evaluated, according to its expected visibility. Roads in viewing direction are therefore higher evaluated, than roads running in azimuth direction. The fusion technique is demonstrated on two sub-urban SAR scenes. The results show the potential of the proposed fusion strategy; with the use of two or more views, the resulting road network is more complete and more correct than for each single image.