NA 
EC Ó MMC 
The data set consisted of both satellite imagery and airborne 
  
0.05 - 
  
Attitude [rad] 
  
  
  
-0.05 
| ca. 200m on ground 
-O4l 
-0.15} 
yaw 
-0.2 2 i x . t : 4 : : 
0 10 20 30 40 50 60 70 80 90 100 
Time [sec] 
Figure 2: Attitude time state of GERAIS image, derived 
solely from automatically marked RCPs. 
imaging spectrometer data. The satellite data included a 
SPOT Panchromatic stereo pair and a Landsat TM scene 
(7 bands). The airborne data contained two sequential 
AVIRIS scenes which were concatenated into a single larger 
scene (224 bands), two non-sequential GERAIS scenes (63 
bands each), a TIMS scene (6 bands) and a GEOSCAN scene 
(24 bands). 
In general, the sensor models for all these sensors were well 
known. However, there were differing amounts of informa- 
tion on the platforms associated with the images. This is 
summarized in Table 1. It is interesting to note that even 
when trajectory information was available for airborne plat- 
forms, it was, in general, quite noisy. 
To begin, a DEM and an orthorectified SPOT image of the 
area were generated from the SPOT stereo pair using tech- 
niques outlined in [4]. Only a few manually marked GCPs 
were necessary for each stereo scene. Similarly, using only a 
few GCPs and the SPOT generated DEM, the Landsat TM 
image was orthorectified. 
The orthorectified SPOT and TM images then served as ref- 
erence images for the airborne images. The AVIRIS scene 
was the simplest to orthorectify because its U2 aircraft fly- 
  
  
  
  
Dataset Auxiliary Information 
SPOT PLA | Orbital Elements 
Landsat TM | - 
AVIRIS nominal position & attitude 
TIMS nom. position (no height) & attitude 
(roll corrected raw image) 
GEOSCAN | nom. start & stop position 
GERAIS - 
  
  
Table 1: Auxiliary information associated with imagery. 
ing at 20 km height provided a stable platform. Moreover 
there was a wealth of measured aircraft position and attitude 
information available. Thus, only a relatively small number 
of RCPs were necessary to orthorectify the AVIRIS scene. 
The other images, on the other hand, had large high fre- 
quency components and severe correlation problems. The 
GEOSCAN platform was particularly unstable, whereas the 
TIMS imagery was particularly hard to correlate with either 
133 
TM or SPOT because of radiometric differences. With the 
exception of AVIRIS, all airborne sensors were orthorectified 
using an acquisition model generated by methods similar to 
the extended Kalman filter method mentioned before. 
Using the methods and prototype system described in this 
paper, it was possible to co-register and orthorectify all the 
images with an RMS error smaller than one 20 m pixel size. 
Examples of the geocoded images from each of the sensors 
are shown in Figure 3. 
In conclusion, geocoding is an essential pre-requisite for quan- 
titative analysis. Using all available a priori knowledge and 
organizing this knowledge in an object oriented design, we 
have shown that excellent geocoding results can be achieved 
even for very hard geocoding problems. An object oriented 
design can easily accommodate many diverse sensors and 
platforms. It also makes it easy to add new sensors, plat- 
forms or auxiliary knowledge about the image acquisition 
process. These properties make object oriented technologies 
an ideal candidate for the development of geocoding facili- 
ties. 
ACKNOWLEDGEMENTS 
The authors would like to express their thanks to Bryan Bai- 
ley of the U.S. Geological Survey and the GEOSAT Commit- 
tee for their help in supplying the image data over the Drum 
Mountains. 
References 
[1] Ralph Bernstein. Manual of Remote Sensing, chapter 21, 
pages 873-897. American Society of Photogrammetry, 
1983. 
[2] B.J. Devereux, M. Fuller, L. Carter, and R.J. Parsell. 
Geometric correction of airborne scanner imagery by 
matching delaunay triangles. Int. J. Remote Sensing, 
11(12):2237-2251, 1990. 
[3] Bruce Sharpe and Kelly Wiebe. Planimetric accuracy in 
satellite mapping. In Proceedings of the XVI Interna- 
tional Congress of the ISPRS, 1988. 
[4] Jeremy C. Wilson and Brian C. Robertson. A modu- 
lar approach to DEM extraction. In IGARSS’91, pages 
1415-1418, 1991.