Skip to content
printicon
Show report in:

UMINF 09.18

Plane-Based Close Range Photogrammetric Reconstruction of Buildings

This thesis presents theory and a prototype computer application for photogrammetric reconstruction of textured 3D models of buildings. The application uses bounded planes to represent building facades. The planes are approximated from a reconstructed point cloud and initially bounded by the convex hull of the relevant points. Multiple bounded planes may be combined into complex, composite models. The intersection between two or more planes is used to update the bounds of the corresponding planes. The focus of the thesis has been to create a streamlined operator workflow that reduces operator work time while creating models of sufficient quality. Thus, the main approach is operator-guided automation rather than a fully automatic approach. Of course, subproblems are solved automatically wherever appropriate. Reconstruction results from several buildings of low to high geometric complexity are presented together with the approximate operator work time required for the reconstruction. Furthermore, a time exposure experiment was performed to investigate the effect of the poor lighting conditions common during the winter in northern Sweden. The results show that the reconstruction is sensitive to a combination of three factors: 1) Low-contrast texture in the foreground, 2) low-contrast texture on the building, 3) poor lighting conditions. However, tripod-mounted cameras and sufficient exposure times are shown to alleviate most of these problems. With images of sufficient quality, the total required operator work time, including photography, is in the order of a few man hours per building. The thesis concludes with a discussion on how to improve the robustness of the applications, reduce the operator time, and extend the prototype to work with other primitives, e.g. cylinders, as well as predefined composite primitives.

Keywords

No keywords specified

Authors

Håkan Fors Nilsson och David Grundberg

Back Edit this report
Entry responsible: Niclas Borlin

Page Responsible: Frank Drewes
2020-07-04