International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
laser data segments of unchanged buildings have been 
identified: 
e Generalisation. At the medium scale of 1:10.000 
generalisation is applied to the objects in the topographic 
database. Small intrusions and protrusions in the contour 
of the building objects have been omitted, already in the 
original mapping process. To allow for removed 
intrusions in the database objects, it was checked whether 
the database object would fit inside the dilated laser data 
segment (Figure 3 top). To allow for removed 
protrusions, it was checked whether the eroded laser data 
segment would fit inside the database object (Figure 3 
bottom). The kernel size of this dilation and erosion 
depended on the specifications of the generalisation 
process. This approach allows for larger differences than 
those that could have been caused by generalisation. Still, 
it proved to be effective for the change detection. 
  
  
Figure 3: Dilated laser data segment of a building with an 
intrusion (top left). Generalised database object 
fits inside dilated laser data segment (top right). 
Eroded laser data segment of a building with a 
protrusion (bottom left). Eroded laser data 
segment fits inside generalised database object 
(bottom right). 
  
  
  
e Random data noise. Noise, of course, is present in both 
the map objects and the laser data. The amount of noise, 
however, is much lower than the size of the 
generalisation effects. The differences caused by noise 
can be accounted for by slightly enlarging the 
morphological kernel introduced above. This will 
increase the tolerance in the change detection. 
e Systematic errors. Systematic offsets were observed 
between the location of groups of buildings in the 
topographical database and the same buildings in the 
segmented laser data. Based on the shape and size of 
these buildings one would, however, conclude that many 
of these buildings were not changed. To avoid a detection 
of a change for these events, for each database object the 
optimal alignment with the laser data segment was 
determined. This shift was applied to the database object 
prior to the change detection. 
e Object selection. The mapping catalogue of the map 
producer specifies which objects are to be mapped. In the 
case of used medium scale map, the catalogue specified 
that not all buildings are to be mapped. E.g. only 
buildings larger than 3x3 m should be included in the 
topographical database. It also specified that only those 
buildings should be mapped that are visible from a street. 
Le., sheds behind buildings were not to be mapped even 
if their sizes exceeded the minimum size requirement. 
These kind of mapping rules first need to be applied to 
the building segments extracted from the laser data. 
Otherwise, many “new” buildings would be found that 
should not be inserted into the topographic database. 
6. CHANGE DETECTION RESULTS 
The above procedure has been implemented and tested. This 
section describes the data used in the experiment, the result 
of the building extraction step and the analysis of the 
detected changes. 
6.1 Data description 
The study was aimed at determining the potential of airborne 
laser scanning for the purpose of change detection for the 
revision of the Dutch TOP10vector. database. This database 
was created for usage at a scale of about 1:10.000. The 
building objects have a location accuracy of 1-2 m. 
The laser data was acquired by Terralmaging with an Optech 
ALTMI225 scanner. The data was recorded with an average 
point spacing of 1.4 m. An area was chosen in which many 
buildings were constructed recently. The area contained only 
little vegetation. 
6.2 Extraction of building segments 
All buildings were detected and extracted from the laser data. 
The building segments are shown in Figure 4 together with 
the road centre lines taken from the topographical database. 
A large number of small sheds in gardens behind buildings 
has been detected. The segments shown in red (dark) were 
automatically labelled as buildings "in the second row". For 
those buildings it was assumed that they should not be taken 
into the topographical database as defined by the mapping 
catalogue. 
  
Figure 4: Segments in a part of the DSM that were 
classified as buildings. In red the building objects 
that were labelled as shed. 
6.3 Offsets between the data sources 
In the final step before the actual change detection, the laser 
data segments were optimally aligned with the building 
contours of the topographical database. Figure 5 shows the 
extracted building segments together with both the original 
position of the database objects (red) and the positions after 
the alignment procedure (black). À clearly systematic pattern 
of shifts between the database objects and the laser data 
segments can be observed. However, the shifts are not 
constant. The directions vary and the sizes range from 2 to 4 
m. In the overlay of the laser data with a more accurate map 
Intern 
  
no Sy: 
showr 
used : 
applyi 
error 
detect 
Figure 
6.4 A 
The re 
and 7 
buildii 
Figure 
buildii 
buildii 
type « 
correc 
a 
Figure 
The de 
All ney