International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
such as single points and break lines should be also extracted by 
user manually. 
Determination of conjugate points from PRISM three image 
strips is conducted with the above mentioned matching methods 
after the accurate orientation of these images have been 
determined so the search space can be much more limited, and 
the computation can be speeded up and the reliability of the 
matching results can be also increased. In this section we will 
briefly describe the combination of the feature point matching 
and grid point matching to get more density conjugate points 
for poor information areas in the images. 
2.6.1 Grid point matching: Feature-based area matching 
always failed in poor information areas such as roofs of 
buildings, tops of trees, shadows of mountains and other objects 
in aero images and satellite images since no feature points could 
be extracted. Thus the disparity maps generated by the above 
matching algorithm contain some blank points without disparity 
values. To compensate these missed points, a grid point 
matching method is used here. 
  
Defined grid matched grid 
Matched point / ay 
—-— Cr -—— 7 
r = I f =} 
1 | | I I | 
1 @ ! À 1 ^ ! I Em 
! | | 1 1 T-lod|d - 
Pp b pei a 1 
I 1 1 | e! I \ 1 ) d 
I I I 1 1 
i | | I I Fol -L--769..1 
asie rE TT \ F7 \ ! 
1 1 I I I 9 ! \ \ ! 
e | ! 1 | Ue 
1 I d - > 
FEET 5er r i aen | A 
l I ! le | o à 
I | I | 1 J ! \ ! 
| 190 + 1°: 1 LO 128. 
Image 1 Image 2 
Figure 6. Concept of grid matching. 
  
  
  
Grid point matching belongs to multipoint matching method. It 
is based the matched feature points and tie points mentioned 
above. Figure 6 shows the concept. Assuming the poor 
information area a smooth surface and using the matched points 
around the poor information area to fit a mathematical function 
with least square method, we define a grid area for the poor 
information area by bilinear elements and get each grid point’s 
gray value. 
2.6.2 Combination of feature point matching and grid 
point matching: In this matching process we take the 
relaxation technique to reduce local ambiguity and achieve 
global consistency for grid points. More detail can be found in 
(Baltsavias, 1991; Zhang et al., 1992). 
3. EXPERIMENT RESULTS 
One set of PRISM simulation data generated with SI-250 
imagery in different areas was used to test our approach for 
DEM generation. The test images were taken with SI-250 at 
600m flight height in Kesenuma area with size of 10km x 1km 
in 2003. The terrain surface covers a median-density residential 
area which lies in a hilly landscape with a local elevation 
difference of over 100 meters. Ground resolution of the 
obtained images was about 10cm. To coincide with PRISM 
imagery’s ground resolution, we resampled the SI-250 images 
before the experiment. Figure 7 shows the resampled images. 
Total 9 GCPs were GPS-surveyed and converted into UTM 
coordinate system with standard deviation of 2cm for horizontal 
direction and 3cm for vertical direction. Only 5 GCPs were in 
the three overlap area of the simulated imagery. All GCPs were 
measured manually with our developed software. Additionally, 
several hundreds of pass points points were semi-automatically 
extracted from the three-overlap area. The accuracy of 
triangulation were about 25cm for horizontal and 30cm for 
vertical direction. Using the obtained triangulation results we 
generated DEM with our approach and compared with the DME 
generated from aerial images. Figure 8 shows the compared 
result. 
  
  
  
( c) Backward view 
Figure 7 Generated PRISM simulation images from S1-250 images in Kesenuma area. 
154 
  
  
    
   
  
   
   
   
  
   
   
  
    
    
   
  
  
   
   
   
   
    
    
     
   
  
  
   
    
    
  
   
   
   
   
  
  
  
    
  
  
  
   
  
  
   
  
  
  
  
   
  
  
  
   
  
  
  
International A 
  
  
  
Two sets of F 
were used to t 
compared the 
with the DEM 
more than 93% 
out that the int 
different. The 
From the triang 
the requiremer 
PRISM image 
matching and g 
PRISM data. T 
the combinatioi 
image generate 
Scanner). 
References: 
Baltsavias, E.P 
matching. Ph. 
Photogrammetr 
Chen T. R. S 
Georeference f. 
3" Internatio 
Developments 
Japan, pp. 71-8 
Chen T., R. SI 
Calibration of t 
System, Photo 
69(1), pp.71-78 
Ebner, H. and ‘ 
Using Digital T 
of Photogramm 
11/578-587. 
Ebner, H., W. | 
on point dete: 
Journal of PE& 
Ebner, H., W. I 
on point deter 
using an ex 
Photogrammetr 
458-464. 
Ebner, H., W. I 
MOMS-02/D2 
Journal of Phc 
332-341.