IS 
riminated 
cording to 
ula of the 
eristics of 
ootprints 
ghbouring 
und based 
  
lygons 
omplexity 
shown by 
  
eun 
Complexity 
tour lines 
10st placed 
hapes. The 
, It agrees 
yrospect of 
of contour 
> extracted 
| points of 
from the 
th of TIN 
criminated 
d the ratio 
it building 
n the same 
re selected 
reduce the 
| building 
uneven as 
  
  
When the group has a lot of contour lines, it shows the tall 
objects, and it indicates the possibility of a building. The results 
of this process are shown in Figure 11. The extracted polygons 
with the shortest perimeter are shown by the bold line, and the 
other polygons are shown by the dotted line. 
420 
  
  
  
   
410 + 
  
  
X iml 
Figure 11 Initial Building Footprints Extraction with Similarity 
5.4 Building Footprints Correction 
(a) Straight-line Segment Separations 
Seeing results in section 5.3(c), the extracted incipient building 
footprints have complicated shapes than actual building shapes, 
because the shapes are created by the laser point clouds. 
Therefore, big corners of buildings are found from complicated 
building polygons to be reshaped into simplified building 
polygons. The algorithm is as follows. 
(1) The central points of incipient building footprint (Po) are 
calculated. 
(2) The farthest vertex of polygon (P;) from Py is calculated. 
(3) The farthest vertex of polygon (P5) from P, is calculated. 
(4) The line segment (P;-P,) that is a diagonal line of building 
is generated, and the line segment (P;-P,) separates an 
incipient building footprint into two arcs. An arc of one side 
is extracted due to the line segment. 
(5) The farthest vertex within an arc of one side from the line 
segment (P;-P,) is calculated. When the distance between 
the farthest vertex and the line segment is larger than a 
certain threshold, the new vertex is created as a vertex of 
the rough building footprints. 
(6) The new line segment is created with the additional vertex, 
and the process (5) is performed repeatedly until there is no 
additional vertex. 
The purpose in this paper is to obtain the rough shape of 
building footprints. The results which using threshold is Sm are 
shown in Figure 12. 
  
420 
410 + 
400 + 
        
  
  
      
Su so 
700 720 740 760 780 
X Im} 
Figure 12 Straight-line Segment Separations 
(b) Correction Process using Ortho Image 
In this section, the building footprints are corrected by using the 
ortho image in the same area. 
720 740 760 
X Im] 
  
Figure 13 Ortho Image 
The edge lines are detected by using the canny filter method 
from the ortho image. The results are shown in Figure 14. 
  
420 
% 
410 
aonb 
san 
390 
Ean 
— 
    
   
  
  
  
330 8% 
= £T 7 
mol xoc Von ^ 
00 720 740 
X Im] 
Figure 14 Edge Detection from Ortho Image 
The edge image is clipped by using the line segment in Section 
5.4(a). And the straight line is detected by using the Hough 
Transform from the clipped edge image. The detected straight 
line is shown in Figure 15 by the dotted line. 
Y iml 
8 
ET 
| 
; j N 
8 
  
  
3U 405 70 à MO o5 7 265 MO ME 
xim 
Figure 15 Edge Detection from clipped Ortho Image 
—225—