new tp
photo n-1
bundle block a
new
orientations
Figure 4: Block Adjustment Phase.
(tp = tie point)
new tp,
add. points
photo n-1
new tp,
add. points
photo 0
new
orientations
(tp = tie point)
improved DEM
Figure 5: DEM Update Phase.
4 EXPERIMENTS AND RESULTS
The concept of Multiple-Patch Matching in the Object Space
for Aerotriangulation has been implemented on an Intergraph
workstation. The purpose of this implementation is to check
the feasibility and effectiveness of the proposed ideas. This
section presents the experiments that were performed and
discusses the results.
4.1 Data Sets and Reference Measurements
Six data-sets were used for the experiments (see Table 1)*.
Each of the first three data sets, OSU, SWISS-2 and WY
contains only one stereo model. Using such sets is import-
ant in order to check the effect of the object space approach,
without involving multiple-patch matching. The other three
data-sets, SWISS-3, TEXAS and OEEPE contain areas that
are covered by more than two images, and therefore multiple-
patch matching must be employed. Images from all data-sets
are of relatively large scale. At such a scale, most matching
procedures face problems because of foreshortening and sur-
face discontinuities. These problems are less acute with small
scale images where a solution is obtained more easily. The
scenes described in the images vary: OSU and SWISS show
heavily structured urban areas; OEEPE shows rural areas and
*SWISS-2 is a subset of SWISS-3. When "SWISS" is mentioned, it
refers to both SWISS-2 and SWISS-3.
402
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
Table 1: Image configuration of the data-sets.
# of # of % Image
photos strips overlap scale
OSU 2 1 60 1:4000
SWISS-2 2 1 60 1:2500
WY 2 1 60 1:6000
SWISS-3 3 1 60 1:2500
TEXAS 6 2 60/25 1:4000
OEEPE 8 2 60/50 1:4000
Table 2: Point configuration of the data-sets.
# of Coverage (images) +00 | Rsd.
poisse]: 2e La 2 1.4 1. 5.1 6 [um]
OSU 24 24 | 7 {mlb 2 2
SWISS-2 24 28.1. ml hadi 3 4
WY 24 924 | — |.—-1-——1-— 2 3
SWISS-3 36 25a] il. TT ES 4 7
5
TEXAS 68 39. 1.13 1.120 À 3
OEEPE 109. 1.42 1:122 1 24 15.1 16,1]. 12.4 40
suburbs; TEXAS shows an air-field (runways, no airplanes or
vehicles); and WY shows a rural mountain area.
In order to obtain both reference (a “ground truth”) and
approximate coordinates, conjugate points were measured
manually for each of the data-sets. In most cases, the loca-
tions of the points were selected at the von-Gruber locations.
Around each location, a few points were measured in order
to facilitate the detection of incorrect matching results. In
Table 2, the first column shows the total number of points
that were measured in each data-set. The next five columns
show the number of points that appear on 2,3,4,5 and 6 pho-
tographs, respectively.
The reference points of the OSU and SWISS data-sets were
measured on a Zeiss P1 analytical plotter, using the original
diapositives. For the WY, TEXAS and OEEPE sets, only
digital images were available. Therefore, measurements were
performed with an Intergraph softcopy station. The available
resolution for the images of the OSU, SWISS and WY sets was
15um, for the TEXAS set-22.5um and for OEEPE-30um.
Quality and consistency of the measurements were checked
by a bundle block adjustment. Since no control points were
involved (as it would be virtually impossible to distinguish
between matching and control errors), seven parameters of
the block were fixed. This involved the identification of the
block coordinate system with the coordinate system of the first
photograph, and setting the scale of the block system to be
approximately the scale of the photographs. The results of the
bundle adjustment are summarized in the last two columns of
Table 2. The accuracy (co), and the maximum residual for
each data set are shown.
As noticed from Table 2, the accuracies of the TEXAS and
OEEPE data sets are somewhat worse than those of the other
sets. There are two possible explanations for this. First,
measuring points across strips without marking them on the
images is difficult. An attempt was made to measure the
points across strips as accurately as possible by creating
Tab
Iterati
Pixel
Wind
# of
Grid i
OSU
SWIS
WY
SWS
TEX,
OEEI
"mode
uring e
possibl
levels (
for the
of one
approx
4.2 |
The m
imatio!
tions tl
AATS)
prelim
adding
from t
thermc
im prov
ematic
problei
ing wi
the ref
For th
pixel s
match
shown
values
OEEP
data-s:
lower 1
with tl
Table.
For ea
the to!
observ
simpli
was di
wrong
In the
pared