Municipal Model Maker

Abstract

Researchers at the University of the Bundeswehr in Munich have developed a deep-learning system to generate 3D relief maps of cities using 2D images generated by synthetic aperture radar (SAR) signals beamed from a satellite. The researchers' SAR2Height algorithm, also trained on airborne lidar or stereo-camera images of the same cities, can render a 3D cityscape–such as this 3D relief map of Munich–with building heights accurate to 3 meters. This way, says University of Houston geosensing systems engineering professor Pietro Milillo, SAR images of a disaster area can rapidly generate data on changes in building height and therefore can yield, he says, “a map of probability of collapse of buildings.”