ildings.
‚ve been
of data
so been
nents in
)perator
as were
ium and
n X and
sible as
Namely,
ygons of
s in the
s and by
»nnected
lated by
fficients
/ on the
1e more
We take
ts as in
hed (in
y.
Renata Jedryczka
We used weights average in calculating absent Z values. The new Z is calculated by
Z=Y w;*Z; where w;=-— 1 n=1..8, depending of numbers Z; <>0 (1)
i-l
If we assume that errors of Z values are numbers from interval (m;,m;) , where m, -minZ,, m, --maxZ, (i-l..n)
i A i
m
then new Z value has error. m «- —-*m,. (2)
m,
Because should be m»zm, then mzm; so error couldn't be much bigger then m; - maximum of errors of neighbouring
Z. Only the newly calculating points are taken into account. The data added from outer sources should also have errors
not bigger then max m».
4 EXAMPLE
The black and white images in scale 1:5000 were used as the test data. They have been scanned with pixels of 15um
that means that the image pixel size was about 0.075m. The accuracy of vector data were 0.1-0.2m."
We usually need to correct much wider regions than the places covered by the buildings themselves so if pixel is
0.075m it is more than enough.
The smallest window in the automatic DEM generation was 21 x 21 pixel, on all the levels of pyramid, consequently
the shortest distances between points in original Z file were about 1,58m. To generate orthoimage in the scale, for
example, 1:2000 on screen with the resolution 100 dpi we should have the Z file with the grid about 0,53 m. The
generation of it started when the original Z file was dense (without holes) and smooth enough . In the Figure 4 we have
DEM for area 197m x 189m corresponding with area in the Figure 1 and 2 . In the interactive process the buildings are
removed.
In this area 26 checked points lying on the ground were selected. They were taken manually from images with the help
of stereoscopic visualisation. The maximum difference in Z values of these points with Z generated in dense file was
1,2, the s, = 0.22 m and r.m.s.= 0.06.
Py
rl, i
A A AN T E eer RAE AR
EUR. TINI Tn, ap, qeu qo i A A
LR E i it A M A A A i A
Ny EC WE Ae
er ie)
we re
d 5
A eA rA We A
WT WRT a TT rr a tr CS
EL ps
e x Peas
As C C Hs
Ag, Tl
AAA AAA AA
re —— ———— i-i E E
Figure 4. Dense generated DEM of selected region with grid size about 12.6 m (corresponding with 8*21 pixels in the
original image).
We received the ground-DEM where grid corresponds with pixel size in orthoimage after bilinear interpolation . In the
Figure 5 we have the visualisation in grey scale of ground-DEM (without buildings) generated for orthoimage in
printing scale 1:2000. Checking the Z values on the 20 draining holes taken from vector data we have the maximum
difference of about 1.5m on 1 point , only 3 points about 1 m and 16 in interval (0.1; 0,6) and s, z0.61 and r.m.s.z 0.2.
" All test data we received from the Aalborg University in Denmark at the opportunity of OEEPE test "Automatic
orientation on data base information" (Jedryczka, 1999).
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 455
