||Projection of Oblique Photos onto 3D Models of Buildings
||Bærentzen, Jakob Andreas (Image Analysis and Computer Graphics, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
||Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
||Construction and texture mapping of 3D city models are usually done using
either a manual or semi-automatic approach. The increasing use of 3D building
geometry in different environments, demands for a fully automated approach
which can lower the time it takes to create a textured 3D city model and thus
lower the cost.
This thesis investigates the use of projective texture mapping as a fully automated
approach to texture map 3D building geometry using oblique photos.
The data consists of simple 3D building geometry and oblique photos with corresponding
Due to the simplicity of the geometry and minor deviations in the accuracy
of the orientations of the oblique photos, two approaches are proposed for improving
the result of the texture mapping. An active shape model is used to
locate the roof section of a building in an oblique photo. The geometry of the
building is then translated to match the oblique photo or the projected texture
is translated to match the building geometry. Another approach is to use viewdependent
texture mapping to dynamically texture map the geometry by only
using the oblique photos closest to the position of the observer. The results are
visualised using a real-time application which allows the user to interactively
explore the texture mapped buildings.
Adjusting the oblique photos to the building geometry using the approach of
the active shape model, gives the overall best results. When using the viewdependent
texture mapping, the results show that the accuracy of the building
geometry becomes less important.
Conclusively, projective texture mapping is definitely a valid method for performing
automated texture mapping of 3D building geometry using oblique photos
- though it requires fairly accurate data.
Creation date: 2008-10-09
Update date: 2008-10-09