`
Initial Interaction
Initial Interaction
First the user classifies the detection problem: dark areas, poor
contrast, occluded buildings, occluded shadows, partly detected L or
T-buildings (this list can be extended). The classification scheme
depends on the performance of the automated system. This qualitative
information is very useful to constraint the search for new
hypotheses. Although the classification is in general not unique
(usually several problems occur at the same time), it is unlikely that
a correct hypotheses will be rejected as long as the classification is
correct.
The second qualitative step is a very rough localization
of the missing building. This can be, for example, any point on the
roof (it is possible to automate this step by clustering rejected
hypotheses, see below). After the initial interaction the most likely
hypothesis can be established by the additional information provided
by the user.
Click on the boxes below to get example images or MPEG-movies
(Note that for the box Initial Interaction there is only an
MPEG-movie describing the localization of a building. As mentioned
above, also the classification of a detection problem is part of the
Initial Interaction. But there is also a movie showing this.)