ınbul 2004 
nates À — 
oints. The 
.13 m and 
> standard 
that of the 
N 
  
lors 
to correct 
trol. point 
the center 
he easting 
rat of the 
z 
  
  
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
2.7 Evaluation of vertical accuracy 
The average of vertical differences between coordinates A and 
coordinates B is 1.38m and the standard deviation of that is 
1.42m. Fig. 7 displays the error as a gradation map. It shows a 
south-north tilt of —1.90m to 2.90m. 
  
N 
   
    
  
  
al 3.00 
Fig.7 Vertical errors 
  
Fig. 8 is a gradation map after a tilt correction of the whole 
model using 3 points on the coastline as GCPs to perform of 
the correction without a ground survey. We obtain a good result 
that the average error is 0.33 m and the RMS is 0.44 m. 
  
N 
€ 
3km 
| | 
-3.00 3.00 
Fig.8 Vertical errors after correction with 3 GCPs taken 
from coastline (e: GCP) 
  
  
  
2.8 Conclusion of the evaluation of 3D model 
In the study, accuracy of QuickBird basic stereo images is 
evaluated and the distortion of the 3D model is limited. A few 
GCP are sufficient when we use those images for topographical 
mapping. 
745 
3. EVALUATION Of DSM 
In this study we evaluated DSM generated from QuickBird 
stereo imagery. 
We generated DSMs of two test fields. One is Yokosuka City 
and its surroundings nearby Tokyo. The other is Central Tokyo. 
3.1 Test field 1 (Yokosuka City and its surroundings) 
This test field is the same as that of the previous study where 
we evaluated the accuracy of the 3D model from QuickBird 
stereo imagery. We use the same QuickBird panchromatic 
imagery. For the evaluation, we used DSM from color aerial 
photographs as reference data. 
3.1.1 Specification of aerial photographs 
The aerial photographs were taken using the RC30 camera with 
GPS/IMU. The date of acquisition is December 2003. It is only 
6 months after the observation of the QuickBird imagery. The 
scale is 1:30,000 and the flight altitude is 4690m. These 
photographs were scanned at a resolution of 20 microns 
(2540dpi) by using a digital scanner. Its ground resolution is 
0.60m. It is almost equal to that of QuickBird imagery. 
3.1.2 Orientation 
In the previous study mentioned above, we confirmed that a 
few GCPs are necessary for the accurate orientation of 
QuickBird stereo imagery. In this study we used five GCPs 
(GPS survey results) and about 40 tie points generated 
automatically, and corrected the RPC model by 3rd order 
polynomials. The result of orientation was very good. The 
RMS of horizontal and vertical residuals at the GCPs and ten 
check points were about one pixel respectively. 
3.1.3 DSM generation 
We generated DSM from QuickBird stereo imagery and aerial 
photographs respectively (Fig. 9 and Fig. 10) by using a digital 
photogrammetry software (LPS). The area is Skm*5km and 
located in the middle of the QuickBird stereo pair scenes. The 
elevation range of this area is about 200m. The grid interval of 
both DSMs is 5m. Therefore, the number of points is 1 million. 
    
Fig. 9 DSM from QuickBird 
  
P