Bitelli, Gabrielle
Y (mr)
ese ome vds
5 B
1
€
SHIRES tot (4. 00 4 00000000
*
*
%
T
E usar s uo
"4| A -4000 to -100
Rd MAN
*
S 3
$
PES AES SESS =
N
CCT TE
HE SEIN ANE
N 5
$
*
A 100 to -50
À 50 to -20
-20 to 20
. C T
5000 10000 12000
8000 10000 P 20 to 50
Xm) @ 50 to 100
(mm) @ 100 to 2000
-2- si [Ye
Figure 3 — Differences between DSM generated with analytical plotter and automatic matching: a) spot size = 7 pixels
b) spot size — 21 pixels.
The classes of the residuals corresponding to the extreme values of the differences are relative to the shadow zones,
with poor stereoscopic view and in correspondence of sudden changes in object depth.
The difference graphs reveal furthermore that there are numerous blank spaces due to the absence of data. Such gaps art
caused by various factors: the correlation coefficient used (Cc=0.7) produces that not all points on the grid are plotted
but only those with Cc 2 0.7; the hidden zones were not plotted by the operator in the analytical environment and, as:
result, were excluded from the comparison. In many cases, the blank spaces coincide with empty areas on the objed
surveyed (windows, doors).
Another test on the facade was carried out with the aim of checking if and how automatic matching is improved by
introducing, in addition to the 334 basic points manually plotted according to a regular grid with side 50 cm, furthe
points obtained by vector analytical plotting. Therefore, entering 2606 3D points obtained with manual vectorisation of
the main features of the object, a DSM was automatically created on the entire right-hand part of the facade. A
comparison with the reference surface model did not reveal any significant changes. It was observed that increasing the
manually entered points to a disproportionate degree did not benefit the search for matches: it is sufficient to enter a set
of points distributed evenly over the surface in order to cover all possible depths.
A final test carried out on the facade relates to the matching coefficient. The minimum acceptable value for this
parameter is set before starting automatic matching and is used by the program to carry out data screening at the end d
matching. A correlation coefficient is calculated for each point on the grid; the pairs of homologous points with i
coefficient greater than the limit value set are plotted, whilst those with a lower value are rejected. As a result, the DSM
produced often has areas without points. The use of such limits, therefore, presupposes that all points with high
correlation coefficient values are correct. In the graph in figure 4, the matching coefficient calculated for each point i
represented with a different colour, juxtaposed to the correct planimetric position. Gaps are represented by blank spaces
and, as already indicated, can often be attributed to empty zones or lack of information, but not only this. The matching
excludes points with a coefficient below 0.7: many blank zones can be explained by the absence of these data.
The graph in figure 4 and the graph of point differences (figure 3b) show that not all points with a high matching
coefficient are correct and vice versa. The matching coefficient used as the limit in this case is not an effective
parameter for measuring the quality of the points plotted.
4 ORTHOPHOTO GENERATION
Once the inner and outer orientation of the digital images is known, creation of an orthophoto requires representation of
the surface of the object using faces (mesh) on which the individual portions of the images are projected.
66 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000.
The o
The I:
In clo
whict
discoi
interp
In ord
the ba
of the
functi
Other
widel
triang
possit
Consi
object
orthor
Ortho
object
previo
Ortho]
object
Orthoy
polyli:
An ex
to the
- the n
and by
triangi
- a de
deforn
the ba:
derive
the ort
- durin
plottin
interpc
plottin
create
