On the Use of the Object Oriented Paradigm for Multisensor Geocoding 
Hans W. Wehn and Brian C. Robertson 
MacDonald Dettwiler & Associates Ltd., 
13800 Commerce Parkway, Richmond, B.C., Canada V6V 2J3 
ABSTRACT 
The ability to conduct quantitative analysis with multitemporal and multisensor datasets is critical to the future success of remote 
sensing based resource management and global change monitoring systems. Such analysis needs the data to be co-registered to a common 
geometric frame of reference requiring the removal of complex geometric image distortions inherent in the raw data. This task, referred 
to as geocoding, can best be accomplished by using all available a priori knowledge to model the physics of the image acquisition. 
Given the diverse nature of this knowledge and the need for flexibility to accommodate the multitude of imaging technologies, the 
object oriented paradigm provides an ideal framework for developing multisensor geocoding software. To demonstrate the utility of this 
approach, an object-oriented geocoding toolbox was engineered and successfully used to geometrically register an extensive multisensor 
data set. 
Key Words: Geometric Image Correction, Orthorectification, Sensor Modelling, Object Oriented System. 
1 INTRODUCTION scanner or a SAR sensor). 
e the sensor platform is known to be dynamically con- 
The quantitative use of remotely sensed imagery is key to the strained both for its translational and rotational de- 
success of resource management and global change monitor- grees of freedom; 
ing systems. Such problem domains require the direct com- 
parison and analysis of large multisensor and multitemporal 
datasets. Before such inter-comparisons can be performed, 
the raw image data must be radiometrically and geometri- 
cally corrected to lie in the same radiometric and spatial 
e there are often navigational data such as gyro readings 
of platform attitude or orbital elements; 
e the platform movement during the imaging time can 
d 7 often be assumed to feature constant (or at least well 
omains. understood) altitude, speed and heading; 
The goal of the geocoding process is to precisely register the 
raw imagery to the earth’s surface. This is a pre-requisite 
to any serious quantitative analysis as the results are only 
meaningful if they can be attributed to a specific region on 
the earth’s surface. Only then is it possible to compare or 
fuse the data and make meaningful observations. 
e the relative importance of the platform movement for 
the imaging process is known. For instance, the distor- 
tions introduced by platform pitch typically cannot be 
distinguished from along-track movements of the plat- 
form; 
e the approximate shape of the earth is known to con- 
Geocoding is accomplished in a two-stage process. In the 
form to a chosen datum; 
first stage, a mapping is constructed relating pixels in the 
raw image to geographic coordinates on the surface of the 
earth. This implies the removal of complex geometric dis- 
tortions introduced during image acquisition by the sensor, 
the platform, the viewing geometry and the local topogra- 
phy. The second stage uses this mapping to resample the 
imagery aligning it with a standard grid in the desired map 
projection. In what follows, the geocoding procedure is dis- e absolute information tying a feature in the image to a 
cussed in more detail. known location on the earth. This information takes 
the form of either ground control points (GCP) marked 
by the user, or registration control points (RCP) ob- 
e the region's topography may be known from available 
or derived Digital Elevation Models (DEMs). 
There are two types of a posteriori knowledge: 
2 GEOCODING METHODS tained via correlation with a base image; 
e relative information (without absolute earth location) 
Given the fundamental importance of geocoding for remote tying two or more raw image features together, so- 
sensing, the question arises how geocoding is best achieved. called tie-points (TIP); 
There are two major geocoding methods: image warping and 
image acquisition modelling. The main difference between 
The first geocoding method, image warping, uses only the 
these methods is their use of a priori and a posteriori knowl- s S S pag y 
Ier t a posteriori information to model the geometric distortions 
edge. Before describing the two methods, consider first the in the raw image[2]. Here, GCPs, RCPs, and tie points are fit 
available knowledge. to functions (typically low order polynomials, or a triangular 
There is a wealth of a priori knowledge which describes the irregular network (TIN)) to model the warping from the raw 
image acquisition conditions: image coordinates to map coordinates. 
The second method, image acquisition modelling, uses both 
e the sensor characteristics are known (eg. a linear array the a priori and a posteriori information[l]. The a priori 
130 
knowledg 
acquisitk 
to fine-tt 
to resam 
The latt 
number 
icantly f 
portant, 
thus exp 
RCPs ar 
expensiv 
relation 1 
age acqu 
polynom 
introduc 
extensib| 
age acqu 
RCP err 
the phys 
reduce t 
discarde 
3 0 
Considei 
knowled, 
organize 
The req 
flexible : 
sors, pla 
be easy 
and plat 
stand ar 
Given tl 
vides an 
geocodir 
istics of 
inherita: 
Polymor 
with res 
of geocc 
models, 
sor mod 
Sensors : 
plicatio: 
object c 
the sens 
a functi 
line-pixe 
Rotatin 
the raw 
was acq 
or whet 
craft. 
Inherita 
to the s 
platforn 
and the 
propert 
Sensor 
specific