ace
point
ents.
rs of
ation
ation
hree-
mea-
spect
), the
@.B)
rays
t ori-
e can
oints
1 of a
(See
down
Xc1 = à11X + 312Y t a13Z t a14
| (2)
yc1 = a15X + a16Y t a17Z + 318
for the left image, and in the form
Xc2 = a21X + a22Y + a232 t a24
(3)
x 2 (Xr Yo Zei)
Z /
model space + Á Z Ded en
e
Y Cor 10.0 —
"ans SR
Ce
X object space
Figure-2 : relative and absolute orientation of a
stereopair of affine images
Yc2 = a25X + a26Y + a27Z + a28
for the right one, respectively. The condition that
Equations 1 and 2 are valid for all object points
photographed in common on the left and right im-
ages can be formulated as
ail aı2 913 314 -Xe1l
815 aie 28317 aig -Yeil
a1 a2 a3 a4 -Xc2
a25 3826 327 2828-yc2
- 0 (4)
which is equivalent to the coplanarity condition of
corresponding rays. Under the condition of Equa-
tion 4 we can form a three-dimensional space
(Xm, YM, YM) which can be transformed into the
object space (X,Y,Z) by a three-dimensional affine
transformation having 12 independent coefficients,
Les
XM B1X + B2Y + B3Z + B4
YM BsX + BY + B7Z + Bg (5)
ZM = B9X + B10Ÿ + B11Z + B12
Also, the space (XM,YM,ZM) is equivalent to the
model space. From the results obtained above we
can find the following characteristics of the orien-
tation problem of overlapped affine images:
1) The coplanarity condition of corresponding
rays can mathematically provide four orienta-
tion parameters among the sixteen ones of the
stereopair of affine images, and
2) The one-to-one correspondence relating the
model and object spaces can be uniquely deter-
mined, if four control points are given in the
object space.
GENERAL FREE NETWORK THEORY FOR
AFFINE IMAGES
In the simultaneous determination of both orienta-
tion parameters of overlapped affine images and
coordinates of object points, the basic equations are
given in the form of Equation 1(Okamoto and
Akamatu(1991)). These basic equations can be lin-
earized as
Xc = Xe + XAA1 + YAA2 + ZAA3 + 1°AA4
+ AJAX + A2AY + A3AZ
(6)
Ye = Yc + XAAs + YAAs + ZAA7 + 1°AAg
+ AsAX + AgAY + AJAZ
Setting up Equation 6 for all affine images under
consideration simultaneously, we have a system of
linear equations in a matrix form as
AA =¢ (7)
in which
A : a coefficient matrix of the system of linear
equations
Ax : a vector of corrections to unknowns
€ : a vector of constants.
Without object space controls the matrix A is sin-
gular and its rank deficiency is twelve. In the gen-
eral free network theory for affine images we have
therefore twelve linearly independent vectors satis-
fying the following relationship:
AG = 0 (8)
where G is a matrix constructed from the 12 lin-
early independent vectors, i.e.,
G = (gi, 82: ,12) (9)
g1 2(A1,0,0,0,A5,0,0,0,-X, 0,0)
gl =(0,A,0,0,0,A5,0,0,-Y,0,0)
gl =(0,0,A1,0,0,0,A5,0,-Z,0,0)
ge! =(0,0,0,A1,0,0,0,As,1,0,0)
gl =(A2,0,0,0,A60,0,0,0,-X,0)
ef =(0,A2,0,0,0,A6,0,0,0,-Y,0) (10)
ge} =(0,0,A,0,0,0,A5,0,0,-Z,0)
ef =(0,0,0,A,0,0,0,A6,0,1,0)
gi (45, 0,0, 0,45,0,0,0,0;0 X)
T
210=(0,A3,0,0,0,A;,0,0,0,0,-Y)
el, =(0,0,A3,0,0,0,A7,0,0,0,-Z)
gl.=(0,0,0,A3,0,0,0,A7,0,0,1)
These 12 vectors can easily be found by linearizing
the three-dimensional affine transformation (Equa-
tion 5) and expressed in the form of Equation 10.
Seven of the twelve vectors are related to the
three-dimensional similarity transformation and
the last five vectors pertain to a model deforma-
tion.