The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bib. Beijing 2008 
475 
The accuracy of building’s vertical lines extraction is not high 
enough, because the result of extraction is usually influenced by 
many kinds of factors. Therefore, in order to guarantee the 
precision of building’s coordinates, both the left image and the 
right image are used to calculate the building’s elevation 
respectively, using the method mentioned above. In this way, 
there are two elevations can be obtained for each vertical edge. 
If the difference between these two elevations is small, the 
average of them will be regarded as the building’s elevation. 
Otherwise, if the difference between these two elevations is 
large, that is to say, only the coordinates of building’s base 
obtained from this vertical edge are credible, and the elevation 
cannot be regarded as the basis for change detection. 
3.2 Algorithm in Change Detection for Buildings 
The changes of buildings, occurred in a certain area, can be 
detected by comparing the building’s coordinates calculated by 
vertical edges and the corresponding building’s model obtained 
from three dimensional city models. The detection is divided 
into two parts. The first part is the position change detected in 
planar. The second part is the change detection for elevation, 
and this part is based on the premise that there are no planar 
changes for the building’s position. Finally, the buildings’ 
changes, occurred in this area, are detected by eliminating the 
buildings gradually which are not changed. In order to obtain 
all of the changes in the area, the bidirectional detection is 
carried on when detect the changes in planar. The changes are 
detected both on the buildings’ model based on vertical edges 
and on the vertical edges based on the buildings’ model. 
The flow chart for detection on the buildings’ model based on 
vertical edges is as shown in Figure 3. Take compare between 
the coordinates of buildings’ outlines and the coordinates of 
buildings’ base calculated by vertical edges, and detect whether 
there are points close to the buildings’ base in the three 
dimensional city models. The allowance error of the distance 
is £ . Moreover, some internal structures of buildings (for 
example corridors and verandas) often have different colour 
with the wall, and these will display different pixel values in the 
aerial images. Usually, it is also regarded as the vertical edges 
when extracting. However, these are not characteristic lines 
which compose the building models, and they do not exist in 
the building models. Therefore, vertical edges need to be 
analyzed whether to be contained in a certain building model. 
Find out the nearest building model to vertical edge, and search 
the maximum value and the minimum value in the direction of 
X and Y respectively. Then, rectangular is established with its 
left bottom comer located on (* min , T mm ) an ^ ' ts r '§ht top 
comer located on (jc max , _y max ) • If the vertical edge locates in 
this rectangular, then it is contained in this building model. 
After complete the judgment for each vertical edge, the vertical 
edges are divided into two parts, namely the buildings added 
newly and the buildings which has not be changed in their 
planar position. The further detection for building’s elevation is 
carried on under the precondition that the building has not been 
changed in planar position. Compare the building’s elevation 
with the elevation calculated by the length of vertical edges, 
and the changes of building’s elevation can be obtained 
according to the difference between these two elevations. 
Figure 3 Flow chart for detection based on vertical edges 
It can detect the destroyed buildings for the detection based on 
building models, which makes up the disadvantage of detection 
based on vertical edges. Check every building in building 
models one by one, and the buildings which are not analyzed by 
the detection based on vertical edges are destroyed buildings. 
4. EXPERIMENTS AND DISCUSSION 
In order to confirm the feasibility of the change detection 
method proposed above, pairs of aerial images got in 2007 and 
the building models in Wuhan are used to carry on the 
experiments. The aerial images are got by digital camera, and 
the focal length is 101.4mm. The size of images is 11500 
pixels*7500 pixels, and each pixel is 0.009mm. The building 
models are established using the software named CC-Modeller 
which is exploited by Swiss Federal Institute of Technology 
Zurich. The experiments are carried on using two groups of data 
respectively. One group is the buildings whose structures are 
complex, and another group is herringbone buildings whose 
structures are simple relatively. The same thresholds are set in 
these experiments. S is 0.4, 5, is 3 pixels and the allowance 
error is 1 meter. Figure 4 is part of the experimental result 
which intercepts from the first group. Figure 4a) is the building 
models, 4b) and 4c) are the results which demonstrated the 
vertical edges extracted from left image and right image in 
corresponding images respectively. The black lines in images 
are vertical edges. Because the difference of pixel values, the 
vertical edges extracted from left image and right image are not 
identical. Detect the changes according to vertical edges and 
building models, and the result shows that there are no changes 
in this area.