stanbul 2004 
  
und 
Independent 
n height and 
topographic 
yn horizontal 
Without lost 
higher than 
re are plenty 
the sampling 
is high. And 
een adjacent 
there has an 
on. 
, we propose 
(l) 
t’s height (H). 
and on the 
de surveying 
| points into 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
[Yi 70 )= x Zul (2) 
[0 J 
And then calculate the quantity of projected points of each grid; 
regard the number value as the DoPP of that grid. So the DoPP 
can be presented discretely. According to the definition above, 
we can calculate the value of DoPP of each grid if we select 
reasonable grid size, for example, form grids according to the 
sampling interval designed by scanning system. And the value 
of DoPP has feature as follows (see Fig. 5): 
| 
A2 I em 3 
Pi TERN Bl 4) 
  
| l 
1 ü 0 1 1 
1 1 
1 1 
Figure 5: DoPP on a horizontal mesh grid with a 
building and other objects 
® DoPP has a smaller value and is evenly distributed on the 
surface of the ground; 
® DoPP obviously has a higher value along the boundary of the 
building (the gray band in Fig.5) than in other regions, and 
form continuous band; 
® The DoPP in the inside region of the building is equal to zero 
if without the effect of noise points; 
® The DoPP of an independent object such as lamp pole is 
greater than that of surroundings objects. 
With use of the above knowledge, select reasonable thresholds 
T), T, object segmentation can be carried out according to 
formula 3: 
| >7 building 
DoPP À cT. ground (3) 
else else objects 
Thresholds T1, T2 are determined by formula 2 according to 
practical condition and the purpose of data processing. After 
segmentation, we can get the range images of ground, of 
buildings and of other objects. The data processing, feature 
extraction and modeling. visualization for theme image are much 
easier than for original range image. Especially segmented 
buildings, feature extraction of building's surface outline based on 
the segmented image can be realized, and the vertical feature of 
building through further combination with the value of Z can be 
reconstructed. 
4. Experimental Study 
Two experimental studies were carried out to demonstrate the 
feasibility of the above-proposed method of segmenting 
different objects and of extracting urban building objects. The 
operational workflow for processing range image data is as 
follows: 1) Form a grid mesh with grid size of 0.1m X 0.1m to 
the surveying region; 2) Project all the points in the range 
image onto the grid and calculate the DoPP for each grid; 3) 
Select reasonable DoPP threshold to segment the original range 
image and get the special topic range image. 
4.1 Case 1 
Fig. 6b is the special range image formed by getting rid of 
ground points from the original range image (Fig. 6a), which 
preserves the targets with certain height. Building can be 
extracted and rebuilt by using the height value information of 
points. 
pole 
wall 
  
(a) 
(b) 
Fig. 6 a) Image for testing building; b) Range image as DoPP>10 
937