In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds), 1APRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
5. RESULTS AND DISCUSSION 
5.1 Result from 3D Line Extraction and Facade Location 
3D lines were extracted from eight images looking at the same 
region from different directions and validated by at least six 
images. An example region is shown in Figure 3. In this region, 
most of the extracted lines are horizontal lines and a few are 
vertical, but none are in other directions. This result is 
consistent with the real situation of the region since buildings 
were the only objectives and all the building edges were either 
horizontal or vertical. 
Extracted lines distributed mostly near the top of buildings. 
This is because lines at the lower part were occluded from some 
directions, thus it was not possible to obtain six validation 
images. The dense horizontal lines on some facades were 
probably induced by the complex horizontal structures. 
Figure 3. (a) Original image © Blom looking from south of the 
example region; (b) Extracted 3D lines; (c) Result of 
facade location 
Using these extracted 3D lines, the output of the facade location 
step was the vertical facade planes (Figure 3(c)). The length of 
each facade was determined by the end points of its grouped 
line members, and its height was defined by the maximum 
height of the members. Almost all the dominant facades were 
correctly located except three, which are marked in the dashed 
circles. Those wrong facades have lower heights than the others. 
The 3D lines leading to them probably came from the edge 
between the vegetation and the pavement at the base of the 
buildings. They can be culled either by setting a certain 
threshold of the heights or by the later process of building 
identification. 
In the example area, only one facade was allocated for each of 
the eight buildings (ten in total). The detected facades for the 
four high buildings near the road and the two rectangular flat 
roof buildings behind were facing south. This could be because 
of the complex structures on their south facades. More lines 
were detected from the complex structure, thus more 3D lines 
on the facades were extracted. The failure of line detection on 
facades facing other directions may be due to the low contrast 
in some images. 
5.2 Result from Roof Location 
A histogram presenting Rcc was plotted after coarse plane 
sweeping step for each of the tested buildings. By the histogram, 
rectangular flat roof buildings can be clearly identified from 
others. Then their heights were refined by plane sweeping with 
5cm step length, and later verified by the heights measured 
from LiDAR data. 
5.2.1 Preliminary Type and Height Definition: Examples 
of a rectangular flat roof building and a gable roof building are 
shown in Figure 4. For each of them, a histogram of the 
matching ratio Rcc on both sides w'as plotted. As shown in 
Figure 4(c), there was a high peak on side one, identifying a 
rectangular flat roof, but no obvious peak on side two due to the 
lack of building. The histogram from the gable roof building 
(Figure 4(d)) shows an irregular pattern with no clear peak on 
either side. Therefore, the histogram was able to identify the 
type and location of the rectangular flat roof building. 
Figure 4. 
tc) 
Id) 
(a) & (b) are original image © Blom with overlaid 
facade hypotheses (dashed line); (c) is the Rcc 
histogram of building with rectangular flat roof in 
(a); and (d) is the Rcc histogram of building with 
gable roof in (b). 
General result was shown in Figure 5 for all the 17 tested 
buildings. Not all the flat roof buildings among them had the 
homogeneous roof plane as the one in Figure 3(a). Some of 
them had different colour patches or textures on top. one with 
many small dorms on top and the ninth building in the test 
having four small towers on the roof. Moreover, the tested 
buildings also contained some ones having big ratios of length 
to width. 
As shown in Figure 5, Rr values of all flat roof buildings with 
rectangle shape were much higher than others, so that they can 
be successfully identified. According to the test data, Tr can be 
set from 3.5 to 5, but we chose 5 due to the preference of a 
robust detection on target building type. The wrongly excluded 
rectangular flat roof buildings should be able to identify in the 
further detection on the left buildings. 
254