Semiautomatic Extraction of Building Outlines from Large-Scale Aerial Images

Abstract

One of the most challenging problems in digital photogrammetry today is automatic extraction of cartographic features from large-scale aerial images. Automatic detection of such featuresh general, and bu?ldings in particular, will iignifi1 cantly improve the map compilation process as well as other I tasks, such as DEM and orthophoto generation. This paper presents a semiautomatic approach for the extraction of buildings from large-scale images. The main goal is to recover three-dimensional (30) outlines of buildings. The extraction should be sufficiently accurate, reliable, and efficient in order to comply with the high standards set by the photogrammetric community. In order to obtain a 30 description, the approach comprises stereo analysis for extracting building heights, along with a general knowledge of building geometry and shadow information. The paper explains the approach, presents experiments, and discusses the results.