= the crosses of the reseau grid
— dirt within the images
Although the crosses have been removed by an image
processing step, their influence could not be supressed
completely, resulting in a misinterpretation of the image
content for object points in the vicinity of crosses. In
average 0.9 % blunders were in the data.
GENERATION OF AN ENTIRE OBJECT MODEL
Data accumulation. Each matching process generates
an individual point grid representing an individual part of
the volume surface. All these local point grids have to be
transformed into the global coordinate system, in order to
allow an accumulation. The transformation is performed
by three rotation and shift parameters, chosen with the
inverse values as used for the transformatign of the
orientation parameters into the local model system.
Due to the chosen image configuration, the evaluation of
the whole surface needs at least 20 stereo models and
can be extended to 25 models if desired.
A look at the image configuration makes obvious, that
adjacent models have a considerable overlap.
Consequently some surface parts are determined at least
twice. These parts need a special investigation, because
of
— the necessity to average close positioned points
— the possibility to reject failures.
As opposed to standard applications with Z(X, Y) surfaces
the data is irregular distributed in the X, Y,Z space, what
requires a special data structure allowing for a sorting and
comparision of similar points.
Data structure. The first step evaluates the maximum
extensions of the data and constructs a data cube which
is divided into individual cubic voxels of a given side
length. The side length defines the snap distance for the
surface
object points
Figure 7: Subdivision of object space into voxels
decision whether points are belonging to the same object
point. In addition, the side length is responsible for the
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
morphologic precision and the amount of data used to
represent the object surface (cf fig.7).
All points found to be within a single voxel will then be
compared, in order to find a decision, which data has to
be kept or rejected. The rejection of points will occur, if a
point is identified as failure, which is based on the quality
measures provided by the matching process.
Furthermore this quality measure may serve as weigthing
information within the averaging step.
As second step, the determination of attribute values will
be done. These attribute values serve as additional
informations to be used in further application steps taking
the data for evaluation or visulization purposes. Typical
attribute values are
= the standard deviations in X,Y and Z for the averaged
point location
= the average angle for the intersection of the image
rays
= the number of participated points
= the identifiers for the images used
The first three values express geometric qualities, the last
one establishes a relation between each object point and
the images involved, what is useful for visulizations.
face | edges
F, [51251 Sa
F, |Si2 S13 Sos
edge | nodes
S15 1 2
$13 1 3
S18 1 8
node | coordinate node attribute values
1 eis List nels G1 Ovi: Crass NAGE 14
2 Xe Lara 2 Guns TyzsC 72110 IMAGE;
3 XYZ, 3 | Ox3,0y30m,--.iMmage,,
Figure 8: Triangulation of point space and generation of
topology and attribute data
In the third step the compressed data is used to build up
the topologic relations between the object points (cf fig.
8). These relations are established by a Delauney
triangulation realized in the 3D space. As result of such a
triangulation, all object points which have to be
considered as neighbours will be selected. To all of them
an edge can be introduced, follwed by triangles which are
constructed out of adjacent edges completed by the edge
connecting their end points. Such a triangle surrounds a
small face approximating the object surface. Taking all
faces, the description of the volume surface is complete.
Using index tables the relations between points with their
attribute values and the other primitives are stored and
may be used in further steps.
Results. As example of an accumulation process the
combination of three adjacent point grids is presented.
The grids are evaluated from the stereo models just in
front of the
about 49.00
AOD
Y A
CN COQ
«
US t,
IRE
NIT
WE
HH
he
IRL
500
!
J
=
ZZ
ASSIS
o UP
Figure 9: |
The accum
of 1 mm re
Fig.9 show
constructec
In order tc
matched pi
areas have
is influen(
parameters
here, that
> the si
reliable
> the low
giving
areas 0
Use of ob
complete :
procedure
data set fo
ZZ
FE
ZZ
=
|