This paper proposes a method for the reconstruction of city buildings with automatically derived textures that can be directly used for facade element classification. Oblique and nadir aerial imagery recorded by a multi-head camera system is transformed into dense 3D \udpoint clouds and evaluated statistically in order to extract the hull of the structures. For the resulting wall, roof and ground surfaces\udhigh-resolution polygonal texture patches are calculated and compactly arranged in a texture atlas without resampling. The facade \udtextures subsequently get analyzed by a commercial software package to detect possible windows whose contours are projected into\udthe original oriented source images and sparsely ray-casted to obtain their 3D world coordinates. With the windows being reintegrated \udinto the previously extracted hull the final building models are stored as semantically annotated CityGML ”LOD-2.5” objects.
展开▼
