/, 2012 
Is have been 
'S, which can 
| Figure 7. In 
s are heavily 
iduced by the 
geometrical 
| the building 
spicuous that 
te low PS 
eason for that 
  
   
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
    
Figure 8. Oblique aerial image of test area "a". The different 
facade structures, leading to quite heterogeneous PS densities, 
are clearly visible. 
A close-up of the situation together with the assigned PS and 
the sensors line of sight is displayed in Figure 9. 
It shows quite nicely that most of the PS are generated by 
structures at the facades. This leads to low PS densities in the 
areas marked by the black rectangles. In the case at hand no 
facades of the mentioned building parts are visible, since they 
are occluded or parallel to the sensors line of sight. In fact 
occlusion is a quite common problem in urban environment. In 
many cases PS can be found just at the top of the facades since 
the rest of it is not visible to the sensor. 
4.3 Areac 
Finally, it is important to stress the variety of factors influencing 
the PS density. A good example for that is a trihedral reflection 
mechanism at a facade formed by the window sill, a part of the 
wall, and the frame of the window with just the right orientation 
to the sensors line of sight. If the window is always closed 
during the acquisition of just another image for the data stack, a 
PS is likely to be induced. However, if the window is opened 
once during an acquisition, the PS may be lost. In essence a lot 
of "random" processes decide if a reflection mechanism is 
persistent over the timeframe covered by the data stack. A quite 
nice example for such effect is shown in Figure 10. One part of 
the building complex shows a quite high PS density (around 3.6 
PS per 1000m?) coloured in green. The other part exhibits a 
considerably lower density (0.5 PS per 1000m?) shown in red. 
A closer look at the actual PS distribution reveals, that PS could 
be found at the right part of the building only. The reason for 
that gets obvious in Figure 11, which shows an oblique view 
Es. 
    
    
Figure 9. PS density in test area b. The building parts marked by 
the black rectangles show a quite low PS density because just 
their roofs are visible to the sensor. 
  
Figure 10. PS density for test area c. One part of the building 
(red) shows a very low PS density, while the other part shows 
a quite good coverage (green). The reason for that are 
construction works. 
aerial image (O MS-Bingmaps). A scaffold is visible in the left 
part suggesting ongoing construction works, which certainly 
leads to a loss of all PS at the particular building part. 
5. CONCLUSION 
A work flow aiming at the fusion of PS point clouds with 
building outlines for the purpose of determining the PS density 
per building has been demonstrated. The procedure consists of 
two steps namely alignment of PS and map data and assignment 
of PS to buildings. A simple Iterative Closest Point algorithm 
turned out to be sufficient for the alignment. The 
straightforward assignment of PS to the closest buildings could 
be improved to enhance the algorithms accuracy in dense built 
up areas. In some cases it might be reasonable to check if a set 
of regular shapes (e.g. planes obtained by extruding the polygon 
edges) can be fitted to the PS assigned to one building. 
However, this is quite difficult due to the quite inaccurate 
geocoding of PS and would definitely fail in case of complex 
roof structures leading to an irregular point distribution. 
The map of PS densities is a good tool to get an overview which 
buildings exhibit a sufficient PS coverage for monitoring 
purposes. However, it does not account for the PS distribution 
at the building. For that a matching of the PS to the polygon 
edges is thinkable. The main problem at that is to distinguish 
facade and roof PS which would ultimately boil down to the use 
of a distance threshold. 
A better way would be to use a 3D city model and match the PS 
to bounding faces. 
Finally, the PS density at buildings is quite heterogeneous for 
    
Figure 11. Oblique view aerial image of test area c. The scaffold 
at the left part of the visible facade indicates construction works 
and explains the low PS density for this part of the building.