buildings with heights above the ground from 8 levels to 19 
levels (assuming averagely 3 meter per level). Buildingl was 
about 34 meters high with a major structure on top of it. 
Building2 was about 25 meters high with a major structure 
on top of it as well. And, Building3 was a complex with the 
highest point at 57 meters. Figure 2 shows the area and the 
three buildings used in the experiments. 
3.2 The Experiments and Results 
The experiments were conducted by using LPDS to generate 
3D points to densify the Lidar data sets of the three 
buildings. The TIN models of the existing Lidar data sets 
with 3 meter spacing are in Figure 3, 6, and 9 respectively. 
And, the outputs of LPDS are shown in Figure 4, 7, and 10, 
respectively. These results were generated by using the 
imagery of 0.1 meter ground pixel resolution. Other pixel 
resolutions were tested as well, however, the ones shown in 
this paper represent the best results. At the beginning the 
experiments, images with 0.5 meter ground pixel resolution 
were used and thought to be sufficient meeting the 1 meter 
accuracy requirement. But, the experimental results proved 
that 0.5 meter resolution was just too coarse. The 3D points 
generated for the flat building tops had variation in elevation 
much larger than 1 meter. As already mentioned above, the 
best results came out of the images with the finest ground 
pixel resolution that was 0.1 meter. Of course, the finer the 
pixel resolution, the longer the image matching takes. For 
comparison and quality check purposes, the TIN models of 
the original 1 meter spacing Lidar data were generated as 
well for the three buildings and those TIN models are shown 
in Figure 5, 8, and 11 respectively. 
  
Figure 3. A TIN model of Buildingl generated from Lidar 
data with 3 meter point spacing. 
  
Figure 4. A TIN model of Buildingl generated with the 
densified building points at 1 meter point spacing. 
   
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
Figure 5. A TIN model of Building! generated with the 
original Lidar points at | meter point spacing for comparison. 
  
Figure 6. A TIN model of Building2 generated from Lidar 
data with 3 meter point spacing. 
  
Figure 7. A TIN model of Building? generated with the 
densified building points at | meter point spacing. 
  
Figure 8. A TIN model of Building2 generated with the 
original Lidar points at | meter point spacing for comparison. 
  
  
  
  
  
  
  
  
  
   
  
  
   
   
  
  
  
  
  
  
  
  
  
  
  
  
  
  
   
    
    
  
   
   
  
  
  
  
  
  
   
  
  
  
  
  
   
    
International . 
  
Figure ' 
data wit 
  
Figure 
densifie 
  
Figure 
original 
3.3 An l 
From th 
can obs 
points g 
more ac 
models 
jagged I 
10, the 
building 
building 
with mu 
points n 
also add