Projection.
amera, the
base for the
ys just one
position, the
1age has been
tation for the
ettings of the
a) that could
ttings refer to
ection of the
be avoided if
he automatic
aper follows
ont.
dicular to the
he top.
icular to the
ght.
icular to the
3. APPROXIMATE CAMERA POSITIONING
USING A VIRTUAL CAMERA
General graphics programming packages provide an extensive
set of graphics functions that can be used in a high level
programming language. Such examples are the GL (Graphics
Library) system on Silicon Graphics equipment, Bentley's MDL
on MicroStation PC platform and the graphics standards GKS
and PHIGS+. In these environments customised software
extensions are possible to deal with the virtual camera.
In the proposed system - based on MicroStation PC platform -
by graphical means and MDL programming techniques, it is
possible to “position” and “aim” the virtual camera anywhere
into the 3-D design session space. In this CAD system there is a
virtual camera associated with each view and this camera can be
turned on or off at will. Various settings associated with this
camera allow user to vary the type of lens and the camera’s
exterior orientation, i.e. the position of the camera in current
design session according to design cube and where the camera
is aimed to. The projection used is a camera-view projection,
and the camera position (Xo, Yo, Zo) and orientation (Up-
vector or à, ©, K) for this particular view can be found precisely
without using control points.
The camera can be placed anywhere in the design session,
looking in any direction. In addition, by moving the camera and
its target point progressively through a model, while the
rendered views are saved each time, a "walkthrough"
presentation of the 3-D object is produced.
The whole operational procedure follows the next steps:
First a screen-window segmentation is performed, and then the
placement of both models (CAD and TARGET) is followed.
The placement of the TARGET model is done in red (red
TARGET model) according to provided image co-ordinates
(pixel co-ordinates) for the four facade's vertices. For this
purpose a low-cost image processing software could be used
(e.g. Adobe's Photoshop, Aldus Photostyler, Paint Shop pro).
The placement of the generic CAD model is done in black
(generic CAD model) according to any available measurements
for facade's edges. The facade object must be planar and is
assumed approximately orthogonal.
The design session used has m, cm and pu-positional units
(pixels) as working units, with resolution 1m=100cm and
lem=10 pu.
The virtual camera is placed by default in front of the planar
facade [Fig. 3] in a vertical distance according to the nominal
photo scale.
Yo=S*c (Eq. 1)
Where:
$: the nominal of the photograph.
C: the camera constant.
The CAD-based procedure for camera position approximation,
is based on adjustment methods trying to minimise the
differences between the "red TARGET model" and the
"generic CAD model" whilst the virtual camera is moved into
the design session. For this purpose either an automatic or a
manual approach can be used [Fig. 4, 5, and 6].
3.1 The Automatic Approach
This approach is an empirical one and such as is based on
attempts for approximating camera position and orientation.
It is well known (Patias P., 1991) that the optical-mechanical
rectification procedure has to be performed in three steps.
453
Similarly the empirical automatic procedure follows these steps
in a (semi)automatic way. Three push-buttons are available in
application's dialog box for these steps (see on figure 2 the
Automatic section).
At first, the red TARGET model is magnified so that to
coincide the generic CAD model on two upper vertices [Fig.4]
At second, the generic CAD model is rotated about the X-axis,
so that both down edges are equal in length [Fig. 5]. During this
step some manual operations in generic CAD model scale are
needed in order the two upper vertices to stay coincide.
Finally, the generic CAD model is rotated about Y-axis, so that
both right edges are equal in length [Fig. 6]. During this step
some manual operations in generic CAD model scale are
needed in order the two down edges to continue being coincide.
During this automatic procedure the current values of Xo, Yo,
Zo, à, ©, and K are displayed for monitoring and feedback
purposes [Fig. 2]. When the automatic approach finishes some
good approximated values for Xo, Yo, Zo, c, 9, and x have
been found.
3.2 The Manual Approach
These approximated values of Xo, Yo, Zo, c, 9, and K are used
in the manual approach method as the initial values of the
exterior orientation. Push-buttons control micro-rotations for ®,
®, X and micro-movements for Xo, Yo, and Zo, according to a
user-defined step. These window-buttons allow changes to be
made to the six degrees of freedom for the virtual camera
separately [see on figure 2 the Manual section].
During both approaches the topological differences of both
models are recorded and these differences define the achieved
accuracy of the approximation procedure. When this accuracy is
not adequate, new approximated values are calculated and used
as feedback to an adjustment algorithm until a predefined
accuracy level is achieved. The adjustment algorithm positions
the virtual camera whilst the aim, that is the “red TARGET
model”, is fixed. During this procedure when an attempt leads
to model differences smaller than a given accuracy, the camera
position and camera settings! parameters are displayed and
saved for future use.
The final values of Xo, Yo, Zo, o, 9, and K are actualy
approximate the exterior orientation parameters of the camera
used to take the particular photo. When these values are found,
the rotation matrix elements (Rij) and the rectification
parameters (al, a2, a3, bl, b2, b3, dl, d2) are calculated and
saved for future use.
The camera settings (lens: Field of Vision angle (FoV), focal
length (f) / camera constant (c)) used in this procedure to
control projection are fixed and they could be found empirically
or they are known from the documentation.
It is important to notice that the virtual camera used in CAD
environments has some advantages over conventional cameras.
Hence in virtual cameras:
a) Everything is in focus; no matter how close to or far from the
camera,
b) There are no problems related with depth of fields,
astigmatism, curvature of field and aberrations.
Finally, as a conclusion to this chapter, it could be said that:
"The approaching procedure consists of consequently generated
generic CAD models trying to match the red TARGET model
according to a predefined accuracy".
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
