Bf, 
into 
Ware 
any 
orms 
the 
vely 
the 
ning 
Photogrammetry 
This procedure is for generating the 
Transformation parameters based on rigorous 
photogrammetric principles. 
1) Inner orientation 
The system can recognize fiducial marks 
automatically or interactively and find the accurate 
position. For automatically getting the coarse 
position of fiducial marks, the CHAMFER matching 
method [1] is used. And the relationship between 
scaner system and photo system has been got. 
2) Relative orientation 
The system also can complete the relative 
orientation automatically based on feature based 
method (B&T, Barnard and Thompson method[2]) or 
hierarchical (pyramid) relaxation correlation (HRC) 
method. The processe of the two methods are in the 
following: 
For the B&T method 
operator choicing approximate positions of 
Gruber points 
getting several pieces of Gruber image 
feature points extraction and non-maximum 
compression 
. make the initial estimate of the probability 
of each possible disparity based on the similarity 
of subimage surrounding the points 
iteratively improving the estimates by a 
relaxation labeling technique making use of the 
local continuity property of disparity 
normalized cross correlation for initial 
position 
least square matching for accurate position 
relative orientation computing 
For the HRC method 
getting Gruber image without interfere 
forming pyramid data 
gradually matching from the lowest to the 
hightest resolution level 
least square matching for accrate position 
relative orientation conputing 
3) Absolute orientation 
Unlike the inner and relative orientation, the 
absolute orientation is driven manually using mouse 
for getting position. 
For quality checking, an error analysis is 
generated listing the residual error on points as 
well as the  Root-Mean-Squared-ErrorCRMSE) for 
fiducial transformation, photo-to-ground 
transformation, photo-to-scaner transformation, 
space resection, and check point-to-ground point 
comparison. The individual point residuals are given 
in ground unit. photo scale, and pixel. The operator 
271 
can reject points and make the Process return to 
appropriate step to regenerate outputs and recompute 
errors. So far the AEDAS has built the exact 
relationship among the scan system photo systems 
space model system and ground system. All the 
relationship parameters are in the database. At last 
in this part, the image will be resampled align with 
the epipolar 
DTM Extraction 
Correlation process is a key procedure in the 
performance of the AEDAS. we have developed our own 
algorithm for automatic terrain data colllection 
which is based on normalized cross correlation. For 
the pull-in task our correlation method uses several 
levels of minified images to gradually build an 
increasing accurate Digital Terrain Model. At the 
hightest resolution level, the system gets accurate 
DIM. “in sub-pixel by analyzing just small 
neighborhood around the peak in the correlation 
suface. For the sake of reliability two criterias 
are selected so that correlations over cloud, water 
or other structureless area are rejected. The first 
is to check the peak of the correlation surface. The 
second is the fall-off rate from the peak of the 
correlation surface in a small neighborhood around 
the peak. The two criterias have been found to be 
effective in identifying false peaks. 
The final step for DTM extraction is reformatting, 
converting the photo data to ground space data and 
performs the necessary resampling to produce the 
desired ground space grid. 
DTM editing 
Extensive tools for editing the elevation data are 
also provided within AEDAS. Elevation spikes and 
holes can be edited, as well as regions 
interactively defined by the mouse. These regions 
can be interactively filtered, assigned specific 
values, or given a particular slope. The DTMs are 
then used in the orthorectific process. 
Orthorectification Processing 
Images are automatically orthorctified based on 
the edited DIM. The system will not allow 
orthorectification to be performed on unprepared 
data. 
Quality Control (QO) 
In fact, QC accompanies the process in every 
steps» such as the two criteries in DTM extraction. 
Specialities have been researching a good QC 
approach to the DTM reliability for several Years. 
The AFDAS is using a method to detect and remove the 
gross error. The following is described in detail: 
1) finding the matching point in right image 
using the left image as target image. 
2) getting a patch of image arround the matching