CIPA 2003 XIX th International Symposium, 30 September-04 October, 2003, Antalya, Turkey 
range (depending on the nominal scale of the map base it is 
intended to be adopted). The MIVIS image geocoding is 
therefore a delicate step to pass through; further complexities 
come from the whiskbroom MIVIS sensor model which 
introduces many deformations it’s important to take care of. 
Scene geometry has to be corrected. Usual methodology based 
on simple polynomial approach cannot model such geometry 
especially in a mountain region as the study one is. 
Orthoprojection has to be done to make MIVIS data suitable for 
the subsequent data integration. 
The whole process path falls into four main steps: 
a) Image orthoprojection; 
b) Significant band selection; 
c) Test area selection and image masking; 
d) Image classification and validation. 
Figure 1. Area acquired by MIVIS airborne scanner. 
3. IMAGE ORTHOPROJECTION 
The MIVIS sensor is a whiskbroom one. A rigorous sensor 
model is not easy to describe and ancillary data needed for such 
operation (attitude and position recordings) are often not 
available. In the rigorous model each pixel is acquired with its 
own perspective different from the other ones. Rigorous 
approach based upon collinearity equations is not appropriate in 
this case. Collinearity equations would have to be modified to 
take the dependence of the sensor attitude and position on the 
time into consideration. Such a complex model can successfully 
be substituted by general projecting algorithms. The Rational 
Function Model (RFM) is one of the most commonly used non- 
parametric one. This model relates the image coordinates (>,0) 
and the three dimensional terrain coordinates (X, Y,Z) through a 
ratio of 3 rd order (maximum) polynomials Pi (20 coefficients), as 
shown below; 
e P a (X,Y,Z) 
P„(X,Y,Z) 
p c (x,Y,z) 
P d (X,Y,Z) 
These equations are called RFM upward equations. 
The model has been implemented in IDL programming 
language in order to control the behavior of this methodology. 
It has been found that it is closely related to the geometric 
distortion due to the sensor attitude, to the elevation range of the 
scene and above all to the number and distribution of the used 
Ground Control Points (GCPs). In this particular case image 
orthoprojection was carried out using the complete RFM (3 rd 
polynomial). 57 GCPs were collected to estimate the model 
coefficients. The residuals on the GCPs are presented below. 
ID 
Residual 
(pixels) 
Res x 
(pixels) 
Res y 
(pixels) 
ID 
Residual 
(pixels) 
Res x 
(pixels) 
Res y 
(pixels) 
1 
Check Point 
32 
1,84 
1,84 
-0,01 
2 
3,01 
-2,74 
1,24 
33 
0,74 
0,10 
-0,74 
3 
2,88 
0,87 
-2,74 
34 
2,32 
-1,14 
-2,02 
4 
2,52 
0,80 
2,39 
35 
1,29 
0,13 
1,28 
5 
1,76 
-1,72 
-0,36 
36 
2,86 
-2,71 
-0,89 
6 
5,52 
5,51 
0,16 
37 
2,43 
2,12 
1,19 
7 
Check Point 
38 
1,66 
1,06 
1,28 
8 
0,70 
-0,43 
0,55 
39 
2,01 
1,70 
-1,07 
9 
1,96 
-0,98 
-1,70 
40 
2,16 
2,14 
0,26 
10 
Check Point 
41 
3,11 
0,60 
-3,05 
11 
"Ô38 | -0,23 | -0,30 
42 
0,95 
0,81 
-0,49 
12 
Check Point 
43 
2,05 
-0,49 
1,99 
13 
1,38 
-0,80 
-1,13 
45 
Check Point 
14 
0,85 
0,69 
-0,49 
46 
1,74 
-1,25 
1,21 
15 
3,15 
-2,80 
-1,45 
47 
0,56 
-0,26 
-0,49 
16 
3,47 
3,36 
0,85 
48 
2,94 
2,92 
0,37 
17 
1,71 
1,48 
-0,85 
49 
4,44 
-4,18 
-1,50 
18 
0,69 
-0,62 
-0,29 
50 
4,29 
-2,38 
3,57 
19 
1,03 
-0,90 
0,50 
51 
2,10 
-1,07 
-1,80 
20 
0,72 
0,61 
0,39 
0001 
1,24 
1,18 
-0,37 
21 
1,03 
-0,99 
-0,27 
0002 
1,23 
0,98 
0,75 
22 
0,09 
0,08 
-0,05 
0003 
1,27 
-1,15 
0,52 
23 
1,08 
0,28 
1,04 
0004 
0,94 
0,84 
-0,42 
24 
0,64 
-0,00 
0,64 
0005 
1,22 
-1,21 
0.20 
25 
1,82 
-0,29 
1,80 
0006 
1,02 
-0,95 
0,38 
26 
0,81 
-0,77 
-0,25 
0008 
0,75 
0,71 
-0,24 
27 
2,42 
-2,38 
-0,45 
0009 
1,16 
1,04 
-0,53 
28 
2,70 
2,10 
-1,69 
0010 
3,15 
3,07 
0,73 
29 
0,88 
-0,71 
0,52 
0011 
“òdi 
-0,04 
-0,10 
30 
1,17 
0,27 
1,14 
0012 
0,91 
-0,84 
-0,35 
31 
0,90 
-0,90 
-0,05 
0013 
2,59 
-2,30 
1,20 
Tab. 1-GCPs Residuals 
The calculated total RMSE is 2.10 pixels, 8 m at the ground. 
Further residual investigations have been carried out on 5 check 
points. 
ID 
Residual (m) 
Res E (m) 
Res N (m) 
1 
28,68 
11,52 
26,24 
7 
15,48 
1,12 
-15,44 
10 
-15,6 
-4,36 
-15 
12 
14,16 
-5,76 
-12,92 
45 
10,6 
-10,48 
-1,48 
ü 
16,904 
6,648 
-14,216 
Tab. 2- CPs Residual 
They show a deteriorated situation basically due to the high 
variability of the area height and a low extrapolation capability 
of the model (low GCPs redundancy?). According to the results 
obtained with other simpler models have driven us to accept the 
former; nominal scale mapping our data could be reasonably 
good is a 1:50.000 (geometrically speaking). 
1:10000 Vector map has been overlaid onto the orthoprojected 
image showing, however, a good correspondence (often better 
than the estimated accuracy).