£
"
2
ax ^ P4 (7^2 x, ^) 2p, x) yy.) d
day ^ 2PB4(X7x9) (y7-y9)* p, (r^«2(y-y )?)
In equations (5) and (4) there are
X) Yo coordinates of the
principal point,
X. y measured comparator co-
ordinates of a point,
ky, k^; ks, P1» D5 distortion parameters,
2. 2 an
wo 'e(yey ^
Thus we obtain the resulting corrections for
the measured image coordinates out of
AX = (x-x3)d,, + dx
(5)
Ay 9 (y7y d, * dy
The extended collinearity equations (1)
become
a,x + ayy + a2 + a)
Xt AX =
agX + aft act 1 (6)
a-X +a-Y + a7 t à
yt oy = 2 6 1 S
agX + a4oY * a z+ 1
and they can be used together with the
conditions (2) for the determination of the
11 transformation and the 5 distortion
parameters out of given object space co-
ordinates and measured image coordinates of
control points.
4, DETERMINATION OF THE PARAMETERS AND
THE OBJECT SPACE COORDINATES
The determination of the unknown parameters
out of (6) and (2) leads to a non-linear
least-squares adjustment if enough object
space control is available. We presume small
corrections Ax and Ay due to distortion and
consider the measured image coordinates
x and y as well as the coordinates x. and y
of the principal point in (3) and (49 as ©
constants.
Thus we obtain linear fractional observation
equations out of (6) with the non-linear
additional constraints (2). For n control
points these 2n+2 equations contain besides
the 11 transformation parameters up to 5
distortion parameters. We take three cases
for the determination of ax and ay into
consideration:
case |incorporated distortion parameters
T k
1
IT ki» Kos K4
III ky» ky, Kgs Pgs Py
The initial values of the unknowns for the
iterative least-squares adjustment are com-
puted in two steps. In the first step we
neglect Ax and Ay in (6) and solve the
resulting system of 11 linear normal
equations for the transformation parameters
a. including all control points. With
- 2 2 2
X7 B (ag + 840 * 44 )
: 2 à 2 (7)
Yo: C7 (ag + alg + 8414 )
and equations (5) we determine in the
Second step the initial values of the
distortion parameters again by solving a
linear system of 1, 5 or 5 normal equations.
Now we have the initial values for the
12, 14 or 16 parameters in the non-linear
least-squares adjustment problem which is
solved iteratively by updating x, and y
in each iteration step and by regarding
them as constants as mentioned above.
After this determination of the transfor-
mation and distortion parameters the measured
image coordinates x and y of all points are
corrected by Ax and Ay out of (5).
With these corrected image coordinates the
object space coordinates of all points,
imaged in at least two photos, are computed
in à second non-linear least-squares adjust-
ment derived from (1) as shown in /2/.
5. PRACTICAL TEST
During the photogrammetric control survey /1/
of a plastic membrane covering an open-air
café we took several slides on a cheap
24 mm x 36 mm film (19 DIN) from Quelle
(DM 6.50) with a KODAK RETINETTE 1B
(1:2,8/45 mm) for documentation purposes.
In order to get a first idea of the
accuracy which is obtainable with the
described on-the-job calibration method and
common slides taken with an inexpensive non-
metric camera we restituted four of these
slides, containing enough object space
control, on a ZEISS PSK.
The position of the ! camera stations and the
control points is given in Figure 1.
14
1300 oO o15
12
o16 017
control points
o 10
19
26 ©
240 © 018
021 o 22
o
23 27
025
y
A A a’ A!
camera stations
E ———————À
30m
Figure 1: Situation of camera stations and
control points