supported by radiometric information in the overlapping area
and that reliability and robustness of results is increased and
validated by an independent control.
Approaches for the fusion of independent clouds of points have
been given e.g. by (Besl & Mc Kay, 1992), (Neugebauer, 1997)
and others. Data fusion there is in general a two step procedure
consisting of registration and integration of point clouds, with
registration being the determination of transformation
parameters which merge separate point clouds into a common
co-ordinate system. Registration of 3D point clouds has been
described in detail by (Besl & Mc Kay, 1992) where the root
means square error of distances is used to determine the
unknown transformation parameters. The integration of
overlapping point clouds is the next step which assures, that a
part of a surface is described by separate points unequivocally.
(Turk & Levoy, 1994) show the integration of a surface net by
intersection, (Neugebauer, 1997) proposes to describe object
space by volume elements (voxels) and to sculpture a surface
from a solid volume. Here, no support of radiometric
information has been used for registration or integration up to
now. However, there are first approaches in this direction.
(Gühring, 2001) presents how data can be acquired and
processed considering adequate co-variances in the processing
step and (Sequeira, 1999) besides geometric reconstruction uses
radiometric information for visualisation purposes.
Figure 1. Rendered test surface
2. EXPERIMENT
In order to validate a controlled process of data fusion trough
numerical experiments, two surfaces from close range
measurement applications are used. Both objects represent free
form surfaces of different type which appear in measurement
tasks frequently.
The objects have been acquired by a stripe projecting device
and also by a digital camera. The digital images have been
plotted by means of stereo-photogrammetry and also by means
of a multiple-image approach separately.
The results of a common adjustment of 3D points and intensity
images using different weights for the observations are
presented and interpreted considering improvements of the
procedure and insight into correct weighting of measurements.
2.1 Surfaces
The first object has especially been prepared for control
purposes, cf. figure 1, the size is 100 x 100 mm? and the
material is aluminium. It has a surface precisely defined by
discrete points which were used for the production of the object
in a milling machine. The precision of the production can be
considered superior to the measurement devices used but as
systematic effects will be described later, no assumptions will
be introduced beforehand.
Furthermore the surface is continuous and opaque but isotropic
and is not textured. In order to prepare the surface for
photogrammetric acquisition, a chalk spray has been utilised
and texture has been applied by a projecting device.
The second object is a gravelled tile of concrete and, thus, has a
much more rough and structured surface than the first object, cf.
figure 2. The plotted size of the object is 125 x 125 mm? An
ideal geometric model of this surface should be a real 3D
description, but here only a 2 1/2 D model has been used. From
this, deficiencies of the result are expected. Furthermore, the
texture of the surface has not been improved by projection.
Figure 2. Rendered concrete surface
2.2 Instrumentation
Both surfaces have been measured with the stripe projecting
device ATOS HR of Gesellschaft fiir Optische Messtechnik
mbH (GOM), the digital images were taken with a consumer
camera Fuji Finepix S1 Pro.
The stripe projector is designed for digitisation of free form
surfaces and uses an analogue projector and two CCD-cameras.
Depending on the size of the object to be measured, different
lenses with varying focal length can be used. From this,
distances between points in object space vary from 0.06 to 0.6
mm having an accuracy of 0.005 to 0.05 mm, (GOM, 2002).
The greatest influence on the accuracy of the system is due to
the instability of the projecting device caused by temperature
effects. An object is now measured by projecting unequivocal
patterns on the surface and plotting the reflected patterns which
have been registered by the cameras.
The photogrammetric registration with Fuji Finepix S1 Pro
yielded 8 bit images with 2304 x 1536 picture elements. The
lens had a focal length of 28 mm and was calibrated
simultaneously.
—106—
For tl
have
Furth
resoh
Ont
acqu
4 an
refle
acqu
vas
2.3
For
pho
that