| in 
lent 
| to 
the 
25 
was 
nts 
)SR1 
and 
ion 
late 
ion 
. on 
ing 
£0 
are 
es, 
POT 
RS, 
| is 
ial 
POT 
two 
bit 
low 
ght 
ped 
on 
ost 
ard 
ity 
SRI 
ite 
az, 
ded 
and 
or 
  
  
  
  
No special coordinates transformations are 
required as is usually the case with the orbital 
parameter model. The coordinates of the GCPs can 
be in any map projection system (e.g. Lambert, 
UTM); they are reduced to a local cartesian 
system by defining a new origin, (round coordinate 
values, near the centre of the model); the heights 
of the GCPs are corrected for the earth 
curvature. This software suite is flexible and 
accepts multisensor images, since no orbit con- 
straints are included. The stereoscopic accuracy 
of SPOT images has been widely assessed during 
recent years,especially by the National Geographic 
Institute (IGN) of France (Rodriguez et al. 1988). 
Results of a large scale experiment (31 scenes, 86 
check points) are summarized in table 3. 
  
  
B/Z Raw Residuals Filtered Residuals 
ratio X Y Z X Y 2 
1 8.1 5.5 4.3 3.8 4.2 3.4 
(27°/-27°) 
  
0.5 7.8 17.2 18.3 14.6 | 44 16.7 
(0°/27°) 
  
  
  
  
  
  
  
  
  
Table 3: Influence of B/Z ratio with SPOT 
stereopairs 
The conclusion was that 6 to 8 GCPs are required 
to obtain a reliable modeling with RMSE of 
residuals of less than 10 m in planimetry and 
height. The GCPs were determined by photogram- 
metric method with accuracies better than 3 m in 
planimetry and 1.5 m in height. 
More recent experiments, within a joint French- 
Canadian project, have led to similar results 
(X=6m,Y=5 m,Z=8 to 9 m) and confirmed the high 
geometric precision of SPOT images (Begin 1991). 
Observations were made at ITC (Venkatesh 1992) 
using different models, stereo SPOT and TM/SPOT. 
Part of the results are summarized in table 4. 
  
  
  
Area p/7 No of Check Points 
ratio |points| X Y Z 
(m) | (m) | (m 
France 
SPOT/SPOT| 0.8 29 8.7 6.7 8.8 
TM/SPOT 0.5 26 11.8 2.2 27.5 
Hannover 
SPOT/SPOT| 0.3 33 12.4 9.2 17.2 
TM/SPOT 0.15 24 15.8 |12.9 46.6 
  
  
  
  
  
  
  
  
Table 4: Summary 
The orientations of all four models were done with 
9 to 12 control points; both TM/SPOT models were 
parallax free. The analysis of the check points 
leads to the following comments: 
- the results of stereo SPOT compares well with 
previous experiments 
- the TM/SPOT accuracy deteriorates by about 25% 
in planimetry and about 65% in height, compared 
to stereo SPOT. 
305 
This is a good result for planimetry but indicates 
some systematic errors affecting the heights, 
which is of course not unexpected. Sources of 
these systematic errors can be found in the: 
- different processing levels of the two images 
(TM and SPOT -1A) 
- geometric distortions of the TM image 
- different resolutions of the two images 
- low B/Z ratio 
But a careful visual inspection of the TM/SPOT 
models reveals more serious "anomalies" in the 
terrain relief. 
ANALYSIS OF MODEL DEFORMATIONS 
There is no clear pattern of model deformation 
which could be easily recognised by visual 
inspection except for some flat areas (e.g.canal, 
airport),showing an undulating aspect in the 
north-south direction (fig. la, 1b). 
There is a clear pattern of deformation in the 
Y-direction with a constant width of 500 m, 
although not regular; this corresponds to the 16 
scan lines of the TM image. 
By close examination of some linear features 
running north-south, in some critical areas one 
can observe a pixel shift for a whole scan 
line, but the extent of the deformation within a 
scan line can not be determined visually. 
This problem of pixel shift is apparently well 
known by TM users and often attributed to the 
resampling method (Hill and Aifadopoulou 1990). 
Our TM data produced by EURIMAGE have been 
resampled by the nearest neighbour method. The 
assumption that part of the deformations could be 
attributed to the procedure of image transform- 
ation can not be totally rejected; this aspect 
needs further investigation. In order to get a 
clearer picture of the pattern and the magnitude 
of the deformations four tests sites were selected 
(fig.2) for carrying out: 
- grid measurements, with an interval of 200 m 
- profile measurements,along the axis of the 
scans;the distance between profiles and points 
equals 500 m. 
The same measurements were performed in the normal 
stereo SPOT and the TM/SPOT models. From these two 
sets of observations height difference models were 
computed, using the SCOP software: 
DIFGRID - SPOT/SPOT - TM/SPOT 
DIGPRO = SPOT/SPOT - TM/SPOT 
The measurements of profiles along the axis of 
scan lines required some special preparation; by 
assigning the same DN values for every 16th line 
in the original TM image, we obtained a stripped 
image with all 16 line sets delimited by white 
lines. In this way the axes of profiles in the 
TM/SPOT model were clearly defined and the same 
profiles could be easily measured in the stereo 
SPOT model. Results of the computations are 
summarized in tables 5 and 6 
j 
i 
i 
i 
SE 
ESA ETS 
se 
Er A EA