in at least two
int coordinate.
ng point has to
pull-in range of
els.
g plane closest
is performed to
e program will
this is reached.
late directions,
te intersections
well as regular
for measuring
lore nodes until
be generated to
uch that their
point sets from
isable format
g wheel
] measurement
or speed during
not suitable for
s the interface
lly because the
k. Thus there is
data from the
e point strings,
ter lines of the
trings lying on
measured or
3 planes it i$
more than one
XF files which
2.2.7 Accuracies and Point Density
The typical camera setup to cover an area of 0.8 m x 1.2 m
leads to a distance between the cameras and the object of
apporximately 1.5 m and a stereo base of 0.7 m. The image
scale here is 1:55, i.e. one image pixel covers an area of 0.5 x
0.5 mm? on the object.
The grid generator typically yields a grid width of 10 mm to 20
mm depending on the curvature of the surface. The expected
accuracy of the grid points is of the order of 0.15 mm.
The typical setup for the profile generator is a point distance
along the profile of 3 mm and a spacing between neighbouring
profile planes of 10 mm to 20 mm. The least squares matching
yields an accuracy of the a single profile point in the order of
0.15 mm (Krzystek at al 1995).
2.3 Camera Calibration
The cameras used are non metric cameras, i.e. they do not
have a fixed interior orientation. In opposition to metric
cameras they lack fiducial marks defining the position of the
CCD sensor in relation to the lense system, the focal length of
the lenses are only approximately known and the distortion
characteristics of the lenses have to be determined.
Since the lenses used were developed for photographic
purposes where distortion plays only a minor role, but it is
here much more important to be able to take photographs also
under unfavourable light conditions, hence they have a rather
larger apperture opening. In addition, they have a large
focusing range from close up photography to infinite distance.
Thus, one can expect that such multipurpose lenses must show
greater distortion values than lenses designed and optimized
for a special task.
This requires a thorough calibration of the lense / camera
system. The lense must be calibrated individually for every
desired focusing distance. For the typical task of measuring full
scale motor vehicle models we decided to focus the lense at
1.5 m and to use a fixed apperture at f-stop 11. This results in
enough depth of view, the distance range at which the image is
still sharp, to suit the measurement task.
The lense / camera system is calibrated by taking images from
16 different positions onto a flat calibration field. The
calibration field consists of approximately 1500 round point
markers on a 1.2 m x 1.2 m larger plate. The point markers are
automatically measured by the calibration procedure and fed
into a bundle block triangulation with additional parameters to
model the deviation of the image coordinates from their ideal
positions.
The result of the calibration procedure are the distortion values
and the focal length of the lens which are used by the
measuring process to correct the measured image coordinates.
The calibration should be repeated periodically to guarantee
constant accuracy of the measurements. In addition the
calibration must be repeated if the lense must be set to another
focusing distance or if the camera was accidentally hit, because
this could move the internal lense system.
Since the applied matching algorythms used in the grid
generator and in the profile generator have an internal accuracy
of 1/10th of a pixel or better the calibration of the internal
orientation parameters should be at least of the same order.
Experience has shown that it is possible to achieve calibration
results better than 1 um, which corresponds to 1/10th of a pixel
of the sensor (Schultes, 1996).
479
3 The CAD system Icem SURF
The CAD system described here represents surfaces using
Bezier polygons. Since a complete object cannot be
represented by a single polygon, the object is segmented into a
number of patches which are tied up by continuity constraints.
The definintion of patches is a manual task and requires some
experience by the operator.
Bezier splines can be defined in this system with varying
polynomial degree and with varying size to suit the shape of
the object. The advantages of Bezier splines are that they can
be easily modified on the computer screen by shifting so called
control points and they lead to a reduction of data handled by
the workstation.
There a two different approaches in working with the data
produced by the photogrammetric system. The first is to start
without an existing CAD representation of the model. This
usually happens, if the stylist has begun his work by
constructing a physical model. Here the photogrammetric
system produces the initial data set from which the Bezier
polygons have to be defined.
Figure 3 - Measured profile data
The operator will load the measured data into the CAD system
and start off with nothing but the raw data on screen. It is the
task of the operator to decide where to put the patches. Usually
one would start off at an arbitrary point of the object to define
the first patch. The patch can now be adjusted in polynomial
degree to best fit the given data, in addition is is possible to
extend the size of the patch until it covers an optimal area. A
new patch has to be defined wherever necessary. Once the
complete object is covered esthetic refinements can be made to
the mathematical representation to ensure positional, gradient
and curvature continuity between the patches. This will
however only lead to an initial setup.
The steps carried out so far only ensure that the polynomial
representation fits as close as possible to the measurements.
This does not yet guarantee that certain esthetic requirements
are fulfilled. Very often the protoype from which the
measurements came also lacks these requirements and in
addition there is a certain amount of noise in the raw data
resulting from small measurement errors.
The next step is to force the polynomial representation to fulfill
the esthetic requirements by applying small corrections to the
patches and hereby deliberately deviating from the raw data.
A very powerfull tool is the projection of reflection lines onto
the design on the computer screen. Especially the design of car
bodies requires that these reflection lines form an even and
continous path along the surface. Even the smallest bump in
the surface will lead to a bent appearance of such lines.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
