Qualifier Seminar by Mr. Adam Patrick Nyaruhuma

Dept. of Earth Observation Science

Verification and update of building vector data using oblique images.

Topographic vector data is an important component in modern society for many purposes including
planning, taxation and location based services. To be useful the data is required to be both
complete and correct in terms of location and geometry. In order to preserve the completeness and
correctness, periodic revision of the data is indispensable. In the following, concentration is on
vector data of buildings.

Building revision involves using current raw data sources to check existing datasets for either of
the following: A building is verified in the database as still there and geometrically accurate; a
building has changed or less detail was captured or was captured with errors and has to be refined;
a building is demolished; or a new building formerly not in the dataset is constructed. In addition
to checking errors, refinement may also involve extension of 2D data to 3D.

Updating of building geometry is a difficult task which has traditionally been done by manual
inspection of aerial images or laser data. To reduce the burden, methods have been proposed for
semi-automation. Many existing and new buildings can be correctly detected using already
proposed methods but, in addition, there are many false alarms.

Traditionally, images for acquisition of building information were captured using large frame
cameras. Normally vertical images were captured with side and forward overlap to allow stereo
processing. Today also small frame cameras can be used. This offers the possibility of using
multiple cameras without much additional cost. Both vertical and oblique images may be acquired
at the same time. Other cheaper platforms such as helicopters or unmanned aerial vehicles (UAVs)
are also used.

While we can only see roof features invertical images, in oblique images we can see the whole structure of a building. Although with oblique images we have more occlusion and we can not see all sides of a building in one image, if many views are available, it may be interesting to combine information obtained from different views. Thus, although it is generally argued that it will probably never be possible to fully automate extraction and update of buildings from images, exploiting
new sources such as overlapping oblique images may give better results.

In this research a new method will be developed for automatic verification and update of building vector data. The main idea is to use evidence given through edges and textures in oblique images. It is also proposed to use multiple overlaps in the images for better verification and for updating the existing data.

previous page
more events