Christian Piechullek
dancy. Another result from this comparison was that while with individual images the problems described in points (2)
(3) and (4) could all be found and partly prevented a correct solution to be obtained, in the simultaneous solution with
all three images these problems could be overcome. This finding proves that the simultaneous solution besides offer
more redundancy, is also more stable. The main reason for this stability is scen in the exploitation oË the stereoscopi
correspondence and thus the ray intersection in object space.
4 EXPERIMENTAL INVESTIGATIONS USING REAL IMAGERY |
4.1 Test goal, input data and image pre-processing
The goal of these investigations was to demonstrate the potential of MI-
SFS for surface reconstruction using real images. Sínce no appropriate Image 30
test images acquired directly in digital form were available we used dig- 4
itised acrial film images. Some image pre-processing (see below) was
necessary since no information about the photographic treatment of the
analogue images could be obtained. By comparing the result of the sur-
face reconstruction to an analytically measured reference DTM we
wanted to prove that the proposed MI-SFS method has the potential of Image 32 + ima
accurately reconstructing a surface of constant albedo. Finally the radius : j valu
of convergence of MI-SFS was to be determined in the same way as in Seti. ; ima
the mentioned simulation studies. i + fit.
For the tests three black and white aerial images with an image scale of M
approximately 1:10.000 of an image strip taken with a wide angle cam- 4.2
era were used (for the set up of the images see figure 4). The images
show a homogencous, poorly textured sand dune area in Eastern Arabia. The
The interior and exterior orientation were determined using an analytical Var)
plotter, the images were subsequently digitised with a pixel size of 30*30 ata
um“ resulting in a ground sample resolution of about 0.3*0.3 m^. The : en |
images of the test area (see figure 5) are about 512x512 pixels in size and : ; rE By The
contain very poor texture and only small grey value differences. The Figure 4: Geometric situation of image {act
illumination direction was calculated from the known time of the image acquisition and illumination, real images devi
acquisition and the geographical coordinates of the imaged surface arca. ":
e
Figure 5: Real images no. 30, 31, and 32. In image no. 31 the area being processed is depicted as a white square
The chosen test area lies in one of the sand dune flanks, where smooth surface undulations, but no edges are prese
The maximum height difference within the area is about 10 meters. For this area, a DTM grid of 35*35 heights witha
mesh size of 2.4 m was measured analytically (see figure 6). Each DTM grid height was measured twice and independ
ently, resulting in a root mean square value for the averaged DTM heights of 0.32 m. The geometric and the radiometr¢
model used in MI-SFS were chosen in correspondence with the analytically measured DTM: the grid size was set to 24
m and the size of the object surface element to (0.3 m). Thus, each grid mesh contained 8*8 object surface elemen
The gradients 9r / 0Z and 3c / dZ in image space amounted to about 1 to 2 pixels/m. Duc to the lack of radiometric ground
truth or independent reflectance measurements, no information was available about the reflectance properties of the
imaged surface.
728 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.