tanbul 2004
AT supplies
e Station
rocess. The
ly accurate
s.
10mologous
7 the kernel
*m 9 m
€
EL. —
e
+
imids for a
e following
ie points are
, where the
mer 1986)
n of feature
ophisticated
nts over the
ic tie point
ramid level.
adjustment
; interactive
the internal
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
block adjustment can be started again, without repeating the
automatic process (Sigle & Heuchel 2001).
Having approximate exterior orientation parameters and image
overlaps available, the AAT approach starts with a block
formation based on feature based matching (FBM) and runs
through overview levels to generate a primary point list. The
match with the largest correlation coefficient yields the most
likely homologue point. Feature based matching determines the
correspondence between image features. Least squares
matching requires accurate approximate values (Figure 4). It
refers to the determination of the correspondence between
image area based matching on the similarity of their gray
values. The cross correlation and the least square matching are
the appropriate methods. So that, after the FBM, the collinearity
equations can be used as a convergence based on least square
matching (LSM) and can work more accurately and improve the
success rate of the matching. (Dórstel et al. 2001).
A control points
BE homologous image
patches with
extracted features
preliminary matching in
all combinations
Q tie point areas P(i)
Figure 4: Matching Scheme in the Kernel System
The work by Tsingas represents the first research on fully
automatic digital aerial triangulation (Ackermann & Tsingas
1994). Instead of gray levels interest points are used as
matching entities and the matching method is based on graphs
(Jaakola & Sarjakoski 1996).
In addition, ISAT generates considerably more conjugate points
and point clusters in the areas of multiple overlap areas. Finally,
the produced data are the orientation parameters for the images
and the analysis of results.
2. PERFORMED TESTS
2.1 Preparation of projects
For testing the AAT algorithm, several tests were performed
with the data available. Firstly, a manual solution and an AAT
solution were performed for the same block of the 1:3500
images. Secondly, the comparison of two solutions for the
1:3500 images, using different number of patches in each one,
was carried out. Thirdly, AAT was applied separately to scales
1:6000 and 1:8000. Moreover, the combinations of 1:3500 with
1:8000 image scales and 1:6000 with 1:8000 were also tested.
The most important project parameters were defined as follows:
as coordinate system the Greek System EGSA87 was used, the
average ground elevation was set at 300m, the standard
deviation for the photo measurements 7um, the 6,,,, for interior
orientation 10um, the o,,,, for the relative orientation 10pm and
the Gmax For the absolute orientation 10um.
The main tests performed include the following projects:
I. Manual aerial triangulation with measurement of tie and
control points in images of 1:3500 scale using ISDM.
Autotriangulation adjustment of the 1:3500 scale images,
using a 3x3 pattern of vonGruber point positions with ISAT
3. Autotriangulation adjustment of the 1:3500 scale images,
using a 5x5 pattern of vonGruber point positions with ISAT
4. Autotriangulation adjustment of the 1:6000 scale images
with ISAT
5. Autotriangulation adjustment of the 1:8000 scale images
with ISAT
6. Autotriangulation adjustment of the combination of 1:6000
and 1:8000 scales with ISAT
7. Autotriangulation adjustment of the combination of 1:3500
and 1:8000 scales with ISAT
N
For every project the appropriate data files were created and the
various parameters were defined. They include, among others,
camera, photo, model, control points, altitude information, such
as flying height and average ground elevation, the desired linear
and angular units, the type of imagery and camera parameters
(such as name, lens distortion values, focal length, and principle
point coordinates), the order of images and strips and the
coordinate system for the control points. The essential number
of overviews for every image is created with the Gaussian
method of resampling - full set is recommended - so that the
images may be better manipulated by the software (Z/I
Imaging, 1999).
2.2 Images of 1:3500 scale
The 1:3500 scale images were ordered in two strips, with a
mean absolute flying height of about 1000m. Initially, the
interior orientation process for every individual photo, with the
help of the corresponding module of Automatic Orientation,
was performed.
2.2.1 Triangulations with manual measurements of tie
and control points: As already mentioned, an adjustment of
manual measurements of the 1:3500 images was initially
performed. Tie and control points were measured in every
photo in which they were well defined.
2.2.2. Autotriangulation (using a 3x3 pattern of vonGruber
point positions): For this project approximate values of the
exterior orientation parameters were provided, with an accuracy
of 50m fof the projection centers, while the o, q angles were
considered equal to zero and the k angle was calculated from
the approximate values of the image centres. The initialization
automatically creates nominal tie-points, in object coordinates,
for later use in correlation. These nominal tie-points (in a 3x3
vonGruber point pattern) will be used later by ISAT to generate
. clusters of matched points. Default value for the size of every
597
patch window is 100x100 pixels. As an example of the
matching method, tie points in multi-photo view are presented
below. The user is obliged to control the success of the
algorithm visually.