A fusion concept for road extraction from multi-aspect SAR data

摘要

Automatic road extraction from synthetic aperture radar (SAR) images is regarded as a complicated task. Due to the side-looking geometry of SAR, shadow- and layover-effects often occlude roads in urban- and forestry-areas. By illuminating the scene from different directions (e.g. multi-aspect images), these effects are reduced. But multi-aspect SAR images contain different information and extracted information is not only redundant and complementary, in some cases even contradictory. Hence, multi-aspect SAR images require a careful selection within the fusion step. In this work, we describe an extension of an automatic road extraction procedure developed for single SAR images towards multi-aspect SAR images. A fusion concept based on the Bayesian probability theory is proposed. Before fusion, the uncertainty of each extracted line segment is assessed by means of predefined probability functions learned from training data. As prior information, global context is incorporated.