1999 
of the 
f im 
timal 
high- 
on of 
1ative 
affine 
r lack 
el of 
anies, 
1ation 
>mote 
ses of 
y will 
s well 
netric 
there 
point 
nges. 
ck of 
image 
'aphic 
els to 
S Was 
ner et 
et al., 
uality 
tio of 
et al., 
SPOT 
  
Fraser, Clive 
  
stereo scenes, accuracies at the 5-10m range in planimetry and 10-20m range in height are readily achievable. The 
corresponding figures for 5m resolution imaging systems such as MOMS-02 and IRS-1C/D are close to 5m in planimetry and 
4-20m in height. The accuracy range in heighting is primarily a function of the various base-to-height ratios encountered in 
cross-track stereo scenes. 
Notwithstanding the presence of imaging perturbations, such as atmospheric refraction, that can be expected to influence 
steerable 1m satellite sensors to a greater degree than SPOT or IRS, extension of the experience gained with 10m- and 5m- 
resolution sensors would suggest that ground point determination to an accuracy of a metre or so should be readily attainable 
with 1m imagery, especially in multi-image configurations. Irrespective of whether the final accuracy capability of Tkonos 
and other 1m-resolution satellites is 0.5 pixels or 3 pixels, the impact on topographic mapping and map revision can be 
expected to be significant. Indeed, a 2 pixel ground point accuracy would enable /konos imagery to be employed for 
cartographic product generation to a scale of as large as 1:10,000, which covers most topographic mapping. Given the 
potential for 1m satellite imagery to render aerial photography at scales of smaller than 1:20,000 obsolescent, it is useful to 
consider the available options for 3D metric exploitation of 1m satellite imagery. 
Within the photogrammetric and remote sensing industries there is a wide desire to metrically exploit 1m satellite imagery to 
the maximum extent possible. This implies application of a fully rigorous mathematical model for orientation and 
triangulation, which in turn implies provision of sensor calibration data and, to a degree, prior information on the satellite 
orbit and sensor attitude data. Where this critical data is not available, there is no alternative but to resort to less 
comprehensive restitution models which might be expected to yield lower metric accuracy. As matters stand at this writing 
(February, 2000), users of 1m stereo Jkonos imagery will be denied access to the *camera model” which could hinder optimal 
metric exploitation of the imagery. The withholding of essential sensor calibration data is seen by the satellite imagery 
providers as necessary for the retention of a competitive edge in the provision of value-add services in the high end of the 
metric product market. 
Thus, alternative restitution models will need to be called upon to allow orientation and triangulation of Ikonos stereo 
imagery. In this paper a number of these alternative models are reviewed, and their applicability to 1m imagery is predicted 
based on experience gained with lower resolution push-broom satellite imaging systems. Initially, however, some further 
comments are offered on the accuracy prospects for 1m satellite imagery. 
2 ACCURACY POTENTIAL 
Whereas the specifications for /konos image products state that a 1-sigma ground point precision of as high as 0.9m will be 
attainable, some simulation studies published to date (e.g. Li and Zhou, 1999) point to more modest expectations of 2-3m for 
planimetric and vertical accuracy, the principal factor limiting precision being the accuracy to which the sensor EO can be 
determined in flight by the on-board GPS receivers and star trackers. In the absence of ground control, but with the 
inclusion of measured ephemeris data, the triangulation accuracy falls off to around 12m (Li and Zhou, 1999), though it 
should be kept in mind that this is representative more of uncertainty in absolute position as opposed to a measure of relative 
accuracy within the restitution of the /konos scene. One could anticipate that the ‘90% Circular Error’ of 50m in planimetry 
for the /konos Geo product could well be considerably improved via simple two-dimensional transformation involving a 
modest number of ground control points. 
Consider, for example, the imaging configuration for /konos illustrated in Figure 1. 
Under the assumption of an image measurement accuracy of 0.5 pixel (Gx, = Sum), 
the ground point triangulation precision to be anticipated for the geometry indicated 
by sensor positions L and R, which have a base-to-height ratio of b/h = 1, is oxy = 
0.32m (planimetry) and 67 = 0.67m (height). If this stereo geometry is extended to 
three along-track images, L, C and R, the triangulation precision in Z remains 
unchanged (see also Ebner et al, 1992), whereas the planimetric precision is 
improved to Gxy = 0.25m, or YA of the ground sample distance. In order to recover 
this level of accuracy, however, an optimal orientation model must be utilised. 
  
Figure 1: Imaging Geometry. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 453