2.2 GCP Selection 
Depending on the circumstances, the exterior orientation 
elements of photographs can be obtained from a previous 
block adjustment or provided by an onboard inertial 
navigation sensor. These data are then stored in a sensor 
parameter file for direct access by the differential rectification 
program. If this is not the case, then GCPs are needed to 
compute the sensor parameters. The PICKGCPs module 
serves the GCP acquisition. 
When well-defined points or targets with known ground 
coordinates appear in the image, the coordinates can be 
measured using the on-screen digitizing routine of CARIS. 
These values are then stored in a coordinate file with the 
corresponding ground coordinates. In the absence of known 
ground coordinates, the operator can manually select GCP 
locations in a digital map and the corresponding image which 
are simultaneously displayed in separate windows. 
The operator can check the residual error of the GCPs by 
performing a projective or polynomial transformation of 
specified degree and reject those which are unacceptable. A 
global index based on "t" distribution is provided for gross- 
error detection. It has been realized that inexperienced users 
often encounter problems in the subsequent exterior 
orientation computation caused by insufficient number or 
distribution of the GCPs. In order to provide users with 
information on the geometric strength of the control 
configuration, the I-1 norm condition number of the system 
inversion is provided. This can then be checked against a 
predefined tolerance of the condition number. 
2.3 Exterior Orientation 
For frame camera photographs the space resection program 
CALCAMP (CALCulate CAMera Parameters), based on the 
collinearity equations, are used to calculate the exterior 
orientation parameters. A residual error analysis is also 
included in the space resection module. Both the condition 
number and the global index for gross-error detection are 
provided. Usually the evaluation of the result of a space 
resection with check points is performed in image space. 
That is, the known object space coordinates of the check 
points are transformed into image space, and compared with 
the measured image space coordinates. It has been found 
that the reverse is also desirable, because a comparison in 
terms of object space coordinates is more meaningful for the 
user. This reprojection must of course incorporate the 
known elevation of individual check points. The 
CHCKPHOT (CHeCK PHOTographic sensor parameters) 
module is designed for the evaluation of space resection. 
2.4 Differential Rectification 
The formation of the orthoimage essentially means a pixel- 
by-pixel reprojection from image onto plane area elements 
set at the terrain elevation of each particular pixel. The 
radiometric or gray-scale value of each new pixel is then 
determined. 
There are two approaches to construct the transformed 
image. One is the output-oriented or indirect approach, and 
the other input-oriented or direct approach. In the indirect 
method, the object coordinates of each output pixel are 
defined first, then transformed into the original image 
coordinate (input) system by the collinearity equations. A 
resampling is then performed to obtain the grayscale value 
from the original image at these image coordinate locations. 
The second method works in the opposite way, in two 
stages. In the first stage, for each pixel in the original 
image, the corresponding coordinates in the output image 
coordinate system are computed and the grayscale value is 
carried with this new pair of coordinates. In the second 
222 
stage, an interpolation is performed to get the grayscale value 
for each pixel in the output image. This is called the direct 
method, because the grayscale value of the original image 
pixels is directly transformed. In CARIS/RIX the output- 
oriented or indirect approach is employed. 
The computation requirement can be reduced by adopting an 
anchor point scheme. This means that not all pixels in the 
original image are transformed, but only the selected anchor 
points (Mueller and Sauleda, 1988; Mayr and Heopke, 
1988). This scheme also assumes that each cell formed by 
anchor points consists of a surface which can be modeled in 
the interpolation procedure. 
Another computation saving scheme is to reduce the number 
of multiplications in the transformation procedure. Due to 
the incremental nature of the raster image structure, a 
significant portion of multiplication in the transformation 
process can be substituted with a simple addition. 
2.5 Mosaicking 
Two or more individual orthoimages can be assembled, to 
form a single continuous coverage of an area, by a program 
called JOINIMAGE. It is based on the algorithm published 
by Hummer-Miller (1989). This scheme has the advantage 
that images can be joined along any seam line, including a 
convoluted one, as long as the line does not intersect itself. 
The seam line is specified by operator selected points on the 
screen. Seam points can also be specified in the map 
coordinate system. The user also has to declare which of the 
two images are retrieved in the overlap area.It is assumed 
that the two images to be joined are in the same coordinate 
system. The image is processed line-by-line, thus the 
memory requirement is low and large images can easily be 
handled. 
3. EXPERIMENT 
The digital orthoimage formation scheme in CARIS/RIX 
was tested in a map revision experiment. The test material 
included a 1: 50 000 scale digital map sheet of the Canadian 
National Topographic System (NTS) series and a 1: 10 000 
scale digital map sheet of the New Brunswick Geographical 
Information Corporation (NBGIC). To facilitate the 
evaluation of the results, various features were deleted in a 
copy of each map file to stimulate an out-of-date map. These 
features were then re-established from the same black and 
white aerial photographs used for the original map 
compilation. The photo scale was 1: 40 000 and 1: 35 000 
respectively. 
The digital images were obtained by scanning the paper 
prints in a Hewlett-Packard ScanJet Plus document scanner 
at 300 dots per inch (118 dots per cm) which resulted in a 85 
um pixel size at image scale. The corresponding pixel size 
on the ground was 3.4 m and 3.0 m in the two photographs 
respectively. The radiometric values were recorded in 256 
gray level. Road intersections and other well-defined 
features were selected in the screen display of the digital 
maps as control for the space resection. 
Both maps cover the City of Fredericton and vicinity. The 
downtown area, which spreads along the shore of the St. 
John River, is essentially flat ground at an elevation near sea 
level. There from the terrain has a steady incline and reaches 
an elevation of 130 m at the city limit. 
Change detection was performed visually on the screen in a 
merged map and image display. The new features were then 
traced in the image by freehand cursor control and digitized. 
The features mapped included: highways, major thorough- 
fares, residential streets, river shoreline, power transmission 
  
sta 
Ort 
acl 
col 
aci 
plc