stanbul 2004 
  
ed with the 
comparison. 
  
from Lidar 
  
ed with the 
  
ed with the 
comparison. 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
  
Figure 9. A TIN model of Building3 generated from Lidar 
data with 3 meter point spacing. 
  
Figure 10 A TIN model of Building3 generated with the 
densified building points at 1 meter point spacing. 
  
Figure 11 A TIN model of Building3 generated with the 
original Lidar points at 1 meter point spacing for comparison. 
3.3 An Evaluation of the Experimental Results 
From the experimental results shown in the above figures, we 
can observe several things. First, the densified building 
points generated more accurate building edges and therefore 
more accurate shapes. In figures 3, 6, and 9, all the three TIN 
models of the original 3 meter spacing Lidar data sets show 
jagged building edges. On the contrary, in figures 4, 7, and 
10, the three TIN models generated from the densified 
building points show generally straight and continuos 
building edges. Second, Building3 is a building complex 
with multiple structures and heights. The densified building 
points not only generated more accurate building edges, but 
also added a lot of accurate details to the complex and even 
to the trees next to the complex, which gave a more accurate 
representation of the building complex. Third, when 
comparing with the real I meter spacing Lidar data TIN 
models of the three buildings, we can see the TIN models 
generated by the densified building points lost almost all 
minor structures on the building tops, especially all medal 
chimneys. But, the loss of all chimneys didn’t really make 
any surprise, because they were too thin to be matched. And 
the last, we can see that the building top surfaces generated 
from the densified building points in Figure 4 and Figure 7 
were not as smooth as the building tops generated from the 
real 1 meter Lidar data in Figure 5 and Figure 8. 
Additionally, in spite of the building edges generated from 
the densified building points are more straight and 
continuous than the building edges generated from the 3 
meter Lidar data points, they are still pretty rough. The 
roughness on the building tops and building edges shows the 
limitation of the image matching. Although the existing 
Lidar data made the image matching process in LPDS much 
easier in finding conjugate points and preserving the building 
shapes, the image matching process still faced certain 
problems. All classic difficulties faced by the image 
matching, except the selection of the search window in this 
experimental environment, were still there causing wrong 
matches or missing matches. Those classic difficulties 
include occlusions, fore-shorting problem, building leans, 
and poor textures. 
Besides the TIN models generated to present the 
experimental results, boundaries for the all three buildings 
were digitized manually from the TIN models to examine the 
horizontal accuracy. Three boundaries were digitized for 
each building: the boundary generated by the 1 meter Lidar 
data (in Light Blue), the boundary generated by the 3 meter 
Lidar data (in Green), and the boundary generated by the 
densified Lidar points (in Red). If the Light Blue can 
represent the true building boundary location and shape, then 
the distances between a Light Blue boundary and a Green (or 
Red) boundary indicate the accuracy of the Green (or Red) 
boundary. In general, we can see in the all three drawings the 
Red boundaries were closer to the Light Blue boundaries 
than the Green ones did, which means the densified Lidar 
points were in better horizontal accuracy and completeness 
than the 3 meter Lidar data. However, apparently, there were 
places where the Red boundaries were away from the Light 
Blue boundaries by more than 1 meter. So, for some 
applications or certain accuracy requirements, additional 
editing would be needed or additional data has to be added to 
the densified data to meet the accuracy requirements. 
  
  
Figure 12. The boundaries of Buildingl: Light Blue one was 
generated by the real 1 meter Lidar data; Green one was 
generated by the 3 meter Lidar data; and Red one was 
generated by the densified 1 meter Lidar points. The bar in 
the middle of the drawing represents 10 meter distance.