ECR- 
this 
Over 
hole 
  
with 
; in- 
July 
| the 
new 
tion 
cm 
icks 
  
The inner orientation of the digital PELICAN images is 
ensured through a radial distorsion model provided by IGN. 
Further controls at CNES confirm the fitness of this model 
with RMS residuals of 0.2 pixel (ie 10 cm). 
Instead of undertaking a costly and uncertain aerotriangulation 
on this block of 1000 low angle images, the exterior 
orientation parameters for each image have been computed 
through the optimization of image matching with the 
corresponding original set of APs using the refined digital 
elevation model available from the original reference data set. 
An average of 3000 points per image were used leading to a 
RMS error of 30 cm after 3 iterations. When landscape 
evolutions were too important for automatic processing 
(typically less than 1000 matching points unevenly 
distributed), hand adjustment was done. 
For data management convenience, the aerial images were 
finally subsampled at a pixel resolution of 1.5 metre, 
compatible with the resolutions of SPOTS data. 
The evolution of the landscape between 1997 and 2002 did not 
justify to compute a specific DTM. 
3 -THE METHOD 
3.1 Description of the method 
The method involved during the in-flight commissioning is 
based on the knowledge of the geometry of acquistion of both 
the reference data and the SPOTS images . The reference data 
is associated with its conical inner and exterior orientations 
and an acurate digital elevation model. The SPOTS images are 
geometrically described by their physical model except for 
the accurate viewing direction angles which are simply 
initialized with the values provided by the laboratory 
measurements ; estimated RMS accuracy of these 
measurements is estimated from about 0.3 pixel for HRG 
instruments to up to a few pixels for HRS instruments. 
Finally the spatial resolutions of the SPOTS instruments are 
described through their MTF models. Following is a 
representation of SPOTS HRGs point spread function . 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004 
  
SPOTS HRG point spread function 
Given a SPOTS image and its associated geometric physical 
and resolution models, each airborne picture of the reference 
dataset is projected in the SPOTS focal plane thus simulating 
the conditions of acquisition of the actual HRG or HRS image. 
For the given reference image let Lm be its associated 
geometric model : for each point (Ir,pr) of the reference 
image the ground position of the point is given by LR(IR,pR). 
Let then Ls be the initial geometric model of the SPOT image: 
Ls (Is,ps) 1s the ground position of point (Is,ps) of the SPOTS 
image. The simulation step then projects point (IR,pr) of the 
reference image at position (Is,ps)=Ls -'Lr(IrR,pr) in the 
SPOTS image while the actual position is (lp). 
The simulated and the real image are then compared to each 
other, using image matching tools. 
3.2 Error budget 
The objective of this paragraph is to analyze the budget errors 
leading to the measured difference 
dm(l.p.Iref) =(1s,ps)-(1,p)+Ecor(1,p) 
where Ecor stands for image matching errors. Image matching 
algorithms developped at CNES/QTIS and applied to this 
method give a RMS accuracy of 0.02 pixel for radiometrically 
similar looking images, which is the case here since 
diachronism as been avoided., 
and 
(Is,ps)=(1,p)+ErEF(Iref)+EDEM(1,p)+EPsF(l,p)+EsPOT(d) 
where 
eREr(Iref) stands for errors of the geometric models of the 
reference data sets; residuals in the bundle ajustment of the 
initial data set lead to a RMS error of 0.4 m. For a given image 
Iref of the reference data set, Erer (Iref) is considered a 
constant. 
£DEM(l,p) stands for errors due to digital elevation model used 
to project the reference data into the SPOTS geometry. 
Refined DEM accuracy is estimated to 1m ; it is increased to 
2m to take into account the landscape changes during the 5 
years period from original data set acquisition and SPOTS data 
acquisition. SPOTS data are all acquired with almost vertical 
viewing directions. The reference data set field of view is 20^, 
thus the RMS influence of DEM is estimated to : 
  
20 * 
] 
EDEM (1,p) = 20° f tan x )dx x2mz03m 
0° 
Epsr(l,p) stands for errors due to point spread function 
approximation ; the SPOTS point spread function is very well 
known and its impact is considered less then a few hundredth 
of a pixel, RMS.