s three distinct
1g phase.
asurement
determine
distortion
JCS460 digital
deo cameras, a
od.
either standard
ocusing ring to
ameters of the
> available later
S is specially
ge and feature
ositions to give
nes possible to
m a moderate
righer accuracy
car model
vith all features
k developed by
‚stem. It houses
lour sensor.
ss suitable for
«els are masked
Iters. Thus, the
erpolated green
sked pixels are
nnels. Although
ause this is the
ure on either of
reduction of the
deration of all
atching process
d by a factor of
ical accuracy in
l.
The measuring system introduced here uses the black and
white variant of the CCD sensor, because only in this model
the full resolution of the sensor is available in a single channel
and colour plays no important role for measuring purposes.
An integrated hard disk drive allows for immediate storage of
the taken images inside the camera, as if using film. The
capacity of the disk supplied with the camera is 170 MByte
which corresponds to 26 images. The disk drives are
exchangable standard PCMCIA devices known from laptop
computers, which allow for a disk change once the disk is full.
Larger capacity drives up to currently 350 MByte,
corresponding to 57 images, are freely avaible on the market.
Since most object surfaces do not feature enough natural
texture, ie.variation in dark and light shades, because in
general they are more or less smooth, a slide projector is used
to project artificial texture on the object.
In order to project the texture over the full field of view of the
camera lense it was necessary to equip the projector with a 35
mm lense.
Two cameras are mounted on a variable stereo base with the
projector centered between them. In principle one could use a
projector fixed on a tripod and a single camera roaming around
the object. Vibrations which occur frequently under practical
conditions, however would cause unpredictable movement of
the texture on the surface of the object. Owing to this fact
always two images with common texture are taken
simultaneously, which are treated as a stereo pair by the
measuring software.
The data transfer of the images is achieved using a personal
computer as a relay station to read out the camera and save
them via network onto the workstation. In addition the PC can
also be used to buffer the images from the camera in case the
internal disk drives of the cameras become full and need to be
saved before they can be tranferred to the workstation.
2.2 The Measuring Software
À short overview of the measuring procedure is given here:
* Administration of image and project data.
* Block triangulation and bundle block
adjustment
* The resampling of the stereo pairs into epipolar
image geometry, and building up of image
pyramids.
* Either the generation of grid ordered point
clouds, or the generation of object sections,
sometimes referred to as profiles, or a
combination of the two.
* Conversion of the generated data into a
CAD-readable data format.
* If required, the generation of point clouds or
profiles can be repeated for instance with higher
point density in areas where high detail is
required.
2.2.1 Administration of image and project data
The software introduced here includes a module for the
administration of the input data. This part includes the
declaration of the image file names belonging to a project, the
setup of image pairs that make of the collection of stereo pairs
covering the object, and the setup up of various default
Parameters like image scale and the assignment of calibration
sets to the image files.
477
2.2.2 Block triangulation and bundle block adjustment
Before the automatic surface measurement can start a block
triangulation and a bundle block adjustment must be carried
out to determine the camera positions and attitudes for each
image taken.
For the purpose of determining the orientation parameters of
the images of the object of interest is covered with some self
adhesive markers serving as tie points between the images.
This is necessary, because the projected texture is not identical
between two different stereo pairs. Thus to be able to connect
up images with differing texture, tie points serve as uniquely
identifyable points that can be easily pointed out by the human
operator Some of these tie points are usually measured
beforehand on a coordinate measuring machine to be able to
determine the absolute reference frame of the object, which are
called reference points.
The operator has to pinpoint the locations of the tie point
markers in the images. This is done by indicating the centre of
the marker using the mouse cursor on the computer screen.
The bundle adjustment software used is able to determine the
image orientation parameters from the given absolute
coordinates of the reference points and the image coordinates
pinpointed by the operator on the images. The bundle
adjustment used is Pictran (Bühler and Gründig,1985) and has
the advantage that no approximate camera positions and
attitudes are required for the adjustment.
The self diagnosis of the adjustment program, based on data
snooping, allows for the identification of weak spots in the
point measurements. It is important to note the fundamental
difference between traditional analoguos and digital
photogrammetry here. As opposed to the former it is possible
here, to correct erroneous measurements directly be
remeasuring the image coordinates on the screen, because the
images are readily available on the workstation.
Remeasuring on a film image requires the redetermination of
the transformation parameters between the image coordinate
system and the carrier coordiante system of the machine where
the film is mounted everytime some image coordinates are to
be remeasured. This makes remeasuring of image coordiantes
uneconomical, instead weak points are just eliminated during
the block adjustment until the block reaches the required
accuracy.
In digital photogrammetry the carrier coordinate system is
replaced by the pixel coordinate system, which remaines
unchanged through the life of the image file. Thus remeasuring
image coordiantes can easily be performed by simply going
back into the image of concern and remeasuring the point
position in the image. This usually results in two to three
iterations of measuring and block adjustment calculation until
the required accuracy is reached.
2.2.3 Normalization of stereo pairs
Once the images are oriented the next phase is to normalize the
image pairs that make up a stereo model, i.e. the images are
resampled to epipolar image geometry. This step requires very
little human interaction, but marking the stereo pairs to be
normalized and pressing a start button on the user interface.
The reasons for the normalization are manifold. The most
obvious is that during this step the image pixels are resampled
in such a way that they are corrected for lens distortions.
Another advantage of the epipolar image geometry is that the
image correlation strategy used can be sped up, because a
small image window of the left image of a stereo pair is
matched to its right image counterpart by only searching along
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
