International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV. Part B2. Istanbul 2004 
  
considerable disadvantage is that new buildings cant be 
detected which is essential for updates of existing data set like 
building layers in GIS or 3D city models. 
  
Figure 1: Grey-value coded representation of a nDSM 
(test site Karlsruhe’) 
es fro (0 e 
be, 1 
ee = 
   
Figure 2: Segmented buildings of the scenery shown above. 
different segments have different colour 
4. CHANGE DETECTION 
The term change detection is used in lots of very different 
contexts. À major interest in change detection technologies 
exists in projects which aim at environmental monitoring tasks. 
In this field, the term is often used for methods capable to 
extract differences in the appearance of natural habitats at 
different dates. e.g. the spread and development of species 
occurring in forests (see e.g. Bhattacharyya et. al, 2002). 
alterations caused by human settlement activities 
development of the city of Istanbul, see Altan et. al., 2002) or 
impacts of natural disasters (e.g. Maffra & Honikel, 2002). 
Most approaches do not concentrate on analysing single objects 
(e.g. single trees of a forest) but on indicating changed arcas. 
Some approaches continue with a classification of the changes 
in the sense that the old and new state of these areas is 
classified, e.g. according to land use or occurring species. 
(eig. 
In recent years, other applications of change detection 
approaches arose which are concerned with an object-wise 
analysis, focussing often on buildings. In contrast to the already 
mentioned change detection approaches, the task in theses 
projects is not to analyse the usage or cover of a special area, 
but to rate the state of a special object at a special date in 
comparison to its state at another one. Above all, this means to 
observe its geometry (changes) Such information can be 
integrated in spatial databases for e.g. urban planning and 
management purposes (see e.g. Matikainen et. al, 2003; 
430 
Jung, 2004). Therefore, this is a kind of general, application 
independent data source. But it can be acquired and used for 
specific, present problems, too, e.g. disaster management tasks 
(compare e.g. Steinle et. al., 2001). The application can have 
implications on the used methodologies, especially the degree 
of automation, and the categorisation of occurring changes. 
Nevertheless, there exist three main groups of building states 
which are incorporated in most approaches: mew, missing (or 
not-confirmed) and not-altered (or confirmed). Dependent on 
the specific application, these classes can be further sub-divided 
and other classes may be added. 
For the project which is presented in this paper, the following 
five classes were defined: 
* not-altered buildings 
* added-on buildings 
* reduced buildings 
* new buildings 
* demolished buildings 
The approach developed in this project is automatic, exclusively 
based on height data and object-based, i.e. it analyses those 
objects extracted by segmentation and classification approach 
which was described in chapter 3. The change detection is 
carried out in three main steps: 
1) extraction of buildings that can not be evaluated 
2) extraction and further analysis of objects that do not 
sufficiently match with the objects at the other date to 
identify new, modified and demolished buildings 
3) classification of the remaining objects into not-altered. 
added-on and reduced buildings 
In the following, the individual steps are described in more 
detail. 
4.1 Not-analysable buildings 
In this work, data captured by TopoSys I sensor was compared 
to such of TopoSys II (chapter 2). As TopoSys I sensor is less 
sensitive than TopoSys II a number of deviations could be 
observed (e.g. Steinle & Báhr. 2002). One of these deviations is 
caused by the characteristic of the sensors when weakly 
reflecting material is hit. Using TopoSys Il. such objects do not 
stick out in the derived DSM, whereas they are missing in 
DSMs computed on the basis of TopoSys I measurements 
(Figure 3 to Figure 6 show an example). 
c 
In the Figure 4 - Figure 6 grey-value coded (visualisations of) 
DSMs are shown, which were computed on the basis of laser 
scanning data. The brighter the colour, the higher the regarded 
region. Black coloured areas show lack of measurement data. It 
is obvious, that the dark roofed building was only partly resp. 
nearly not registered by TopoSys 1 sensor. 
This effect is related to the system's ability to differentiate 
between the backscattered measurement signal and background 
noise. TopoSys I did not have a receiving unit capable enough 
to register the poor pulse intensity scattered back by the roof in 
the example, whereas TopoSys II shows a significantly higher 
sensitivity. 
  
Interna 
Figure : 
pul: 
(To 
Buildin 
regarde 
not obs