Isao Miyagawa 
  
Some DEM (Digital Elevation Map) heights in the table were measured by a laser-ranging-system. There were cases 
which buildings heights could not be estimated using this system. The DEM heights in this table were the average values 
Japan, and has from some points on the top surfaces of buildings. 
rth, GPS). The 
ve to sea level, Table 2: Accuracy Estimation 
  
  
  
  
  
  
  
  
  
  
  
  
  
e projected to DEM height | Method A | Method B | Method C 
on method are 26.97 31.72 28.77 27.72 
Yjesiy Z esti 27.97 32.09 29.18 27.47 
27.48 30.34 27.51 26.70 
33.00 39.55 35.38 35.75 
36.16 39.37 35.09 36.16 
tf Pu 37.34 42.26 37.37 39.49 
Prost 32.70 36.38 32.25 34.58 
; 39.51 45.72 40.32 41.74 
pod 36.79 44.50 42.13 40.45 
| d 40.75 47.40 42.10 43.48 
  
  
  
  
  
  
The results show that some of the 3D object shapes acquired by Method B were more accurate than those yielded by 
Method C. We confirmed that the factorization method with hybrid feature points was effective in achieving accurate 3D 
arth plane cap- object shape. 
ple, from south 
or information, 
86). 
   
Method A Method B 
    
Method C 
Figure 7: 3D Reconstruction 
se, on which the 
bject shape. W 
ints. The 3D re 
; paraperspectiv pe SD 
with hybrid fe A comparison with 3D reconstruction on a flat plane and a slope plane based on 3D delta points, is shown in Figure 8. The 
livagawa, 2000} 3D reconstruction on flat plane fails to realize shape recovery of buildings. Though the ground is unevenness, 3D object 
  
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 613