793 
of the resolution of the imagery. Since the swath width 
is limited, it will frequently be necessary to have several 
overlapping flight lines to obtain the necessary coverage. 
The separate flight lines of imagery will then have to be 
tied to each other and to the existing ground control in 
order to build up a mosaic image of the desired area. 
To put the data volume question into perspective, the 
general experience at CCRS has been that each flight 
line generally contains 8 bands of imagery with 20000 
to 30000 scan lines of 1024 pixels each which requires 
approximately 200 Megabytes of disk storage. Thus for a 
project requiring more than 3 or 4 flight lines, a system 
disk storage capacity of 1 Gigabyte or more very quickly 
becomes a necessity if one wishes to have quick access to 
both raw and processed imagery. 
2 Geometric Processing 
The basic concept behind the geometric correction and 
resampling processing is to compute the exact location 
of each input pixel in the output coordinate system and 
then to resample the imagery onto a uniform grid. Before 
this may be attempted, it is necessary to have accurate 
knowledge of the time history of the sensor motion. Al 
though in theory it should be possible to solve for the mo 
tion of the sensor based entirely on ground control points 
and stereo imagery, the accumulated flight experience at 
CCRS for several different types of aircraft has indicated 
that the aircraft motion contains sufficient high frequency 
components to render the concept impractical. Thus the 
acquisition and recording of auxiliary position and atti 
tude data for use in post-processing of the imagery has 
become the chosen method at CCRS (Gibson, 1986). The 
position and attitude data are recorded from an inertial 
navigation system simultaneously with the imagery data 
in a manner which maintains an exact time correlation 
between the two data sets. Although the inertial sys 
tem parameters contain very good relative information 
(i.e. from sample to sample), they are nevertheless cor 
rupted with low frequency errors in the form of position 
offsets and drifts, and an angular misalignment between 
the inertial coordinate frame and the imager coordinate 
frame. Therefore, before the inertial system data may be 
used to compute the position of the imagery pixels, it is 
necessary to compute and remove these inherent errors. 
This is accomplished by identifying control points in the 
imagery for which the ground coordinates are known and 
then adjusting the navigation system data to achieve a 
least-squares fit of the imagery to the control points. 
2.1 Motion Data 
The motion data recorded in the aircraft consists of veloc 
ity, position and attitude data as derived from the iner 
tial navigation system. In addition, aircraft flying height 
data is also recorded from a high accuracy barometric al 
timeter. At the present time, a project is underway to 
incorporate Global Positioning System (GPS) data with 
the existing inertial system data sets in order to reduce 
or possibly eliminate the requirement for ground control 
point measurements. 
2.2 Photogrammetric Solution 
The algorithm for adjusting the navigation system data is 
based on rigorous photogrammetric principles and incor 
porates the standard Collinearity and Coplanarity condi 
tions (Slama, 1980); the former for utilising ground con 
trol points and the latter for tying multiple flight lines 
together using features of unknown location common to 
overlapping flight lines. The condition equations contain 
non-linear terms involving the attitude angles. A least- 
squares solution to the condition equations may be ob 
tained however, by deriving a linear approximation to 
the equations and following an iterative procedure which 
updates the estimated parameters in each step until no 
further improvement is possible. This approach is well 
known for conventional photo-mapping projects. The al 
gorithm was modified to handle the MEIS situation which 
does not have a central perspective point like a camera 
and also to accommodate the time sampled position and 
attitude measurement data from the inertial navigation 
system. The low frequency random errors in the iner 
tial navigation system data are represented in the system 
error model as low order polynomials and subsequently 
removed as part of the adjustment process. 
The following is a brief overview of the derivation of the 
photogrammetric correction algorithm. Refer to Figure 1 
for illustrations of the following vector definitions: 
• The sensor flight path as a function of time consists 
of a position vector p(t) and an attitude vector a(<), 
both specified with respect to the Geocentric Carte 
sian Coordinate frame X, Y, Z. The attitude angles 
specify the rotation components between the Geo 
centric Cartesian coordinate frame and the sensor 
coordinate frame. 
• The sensor position and attitude at the instant in 
which a specific ground feature is viewed in the 
forward channel are pj and a, respectively, that is 
Pi = vi 1 }) and = «(*>)• 
• The pointing vector for any specific pixel in the for 
ward channel is represented in the sensor coordinate 
frame by the vector m, . 
• The pointing vector (i.e. where the sensor pixel is 
looking for any specific pixel in the forward channel 
is represented in the Geocentric Cartesian coordinate 
frame by the vector xj and is related to mj by the 
equation Xj — Cjirij, where Cj is a direction co 
sine matrix derived from the attitude angle vector 
Cj = C(d}). 
• In a similar manner, the equivalent quantities for the 
aft view are obtained by substituting the subscript k 
in place of j in the above definitions.