International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
  
1065011 
  
  
MÀ 
1065010 
  
  
  
   
| 
| 
  
   
Ne 
1063007 | 1064007 | 1065007 
= x 
LLL E RA TT E 
108 1064008 1065008 
Figure 4 a. The valid mosaic polygons in a block area 
  
Figure 4 b. The measurable seamless stereo model 
Figure 4. The experimental results 
Table 3. The error of |MAX| and RMS of terrain points 
  
X (m) 
0.083 
0.044 
Y (m) 
0.074 
0.042 
Z (m) 
0.084 
0.045 
  
IMAX| 
RMS 
  
  
  
  
  
  
  
nne 
  
13 9 4 55 i1 3 39 101 11 42 1344 dc d5 d di onm 
E DY []DZ 
Figure 5. The error distribution diagram of terrain points 
EJ DX 
For Data set II, the twenty houses were selected for testing. The 
heights of these houses have been measured from the original 
photo pairs at the SOCET SET v4.4.0 of Leica Geosystems, Inc. 
The positions of these points are located the corner of the house 
roof. That is, twenty house roof corners are measured in the 
measurable seamless stereo model. The maximum error and 
RMS error are listed in table 4. Figure 6 illustrates the error 
distribution of these twenty points. 
Table 4. The error of |[MAX| and RMS of non-terrain points 
  
  
  
  
  
  
X (m) Y (m) Z (m) 
IMAX| 0.075 0.080 0.105 
RMS 0.046 0.047 0.067 
  
  
  
As it can be seen from Tables 3 and Table 4, the RMS errors are 
very small. It means that the measurement results from the 
measurable seamless stereo model are nearly similar with those 
in the original photo-pair model. Compared with Table 1, it can 
be found that the accuracy of 3D measurement greatly 
improved. 
005 
  
01 
4.15 
12323487568 78 SE TRISTE EU 
EJDX MDPY CIDZ 
Figure 6. The error distribution diagram of non-terrain points 
6. Discussions 
The measurement of 3D coordinates using orthorectified stereo 
images (two orthoimages of the same region (DTM) from each 
of the images of a stereo pair) has been discussed by Baltsavias 
(1996). The accuracy of object coordinate determination from 
the orthorectified stereo images is as high as that achievable 
when using the original unrectified digital images if the 
measuring accuracy in the original and ortho-images is similar 
(which will be the case if the ortho-images pixel size is similar 
to the scanning pixel size). The correct X, Y, Z measurement 
still can be made, even if the underlying DTM is with the 
blunder error (Baltsavias, 1996). 
As the coordinates inverse transformation from stereomate to 
the orthoimage by the parallax function and DTM is a-rigorous 
computation process, the measurement accuracy from the 
measurable seamless stereo model is completely dependent on 
the measurement accuracy in the orthorectified stereo images. 
  
  
  
   
  
  
  
  
  
  
  
  
  
  
  
  
   
   
  
   
  
  
   
   
   
   
    
     
   
  
  
   
  
  
  
  
  
   
  
   
  
    
   
   
   
   
   
     
   
  
   
  
  
   
   
   
     
International Arch 
Ve n 
That it to say, th 
seamless stereo m 
ordinary stereo mc 
is similar with th 
images. Those col 
measurement accu 
the photos and tl 
nothing to do v 
orthoimage and s 
seamless stereo 1 
applications, espe 
measurement accu 
completely error-f 
The experimental | 
with the above acc 
7. Conclusions 
In this paper, the 
model is proposed 
and algorithm for 
are presented. Th 
accuracy of 3D n 
stereo model is € 
methodology, it ca 
is great, i.e. by 20 
Though the stereo 
measurable seaml 
provides the possi 
measurement, ap: 
traditional 3D mee 
an orthoimage : 
measurement is a 
through recording 
orthoimage and m: 
From practical poi 
such a measurable 
integrated into GI: 
practical and ecor 
terrain features in 
accurate three dir 
features visible fo 
and/or other appl 
photo-interpretatio 
electric power, gec 
References 
Baltsavias E.P. , 1€ 
the extraction 
Journal of Phc 
63-77 
Blachut, T.J., 19