Retrieving Structural Information on Anthropogenic Objects from Single Aerospace Images

— A method for the three-dimensional reconstruction of buildings from a single aerospace image, which consists of two stages—the extraction of semantic information and the restoration of the geometry—is described. The topology of artificial neural networks by the semantic segmentation of building components and reference objects is considered. In the second stage, some mathematical transformations are presented: by calculating the photometric parameters of an image based on metadata or reference objects, by converting spatial coordinates into axial and flat image coordinates, etc. Two examples are shown for calculating photometric parameters and a three-dimensional building model from a single satellite image and an aerial photograph.