The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
994
x _ f afX-X si ) + bfY-Y si ) + cfZ-Z si y
a fX-X si ) + bfY-Y si ) + c 2 (Z-Z si )
0 f a 2 (X-XJ + b 2 (Y-Y si ) + C fZ-ZJ '
afX-X si ) + bfY-Y si ) + c 3 (Z-ZJ
(1)
Where a l ,b j ,c i (/ = 1,2,3) are elements of rotation matrix
composed of (p,co,K, which are angle elements of exterior
orientation.
Two observation equations can be constructed with one ground
control point (gcp). So at least 6 gcps and satellite orbit
parameters are needed to achieve elements of exterior
orientation of linear array push-broom imagery.
2.2 F. Leberl Model
The geometry relation between ground and image points in a
SAR image is established by the Doppler and Range equations
proposed by F. Leberl. (Xiao Guochao et al., 2001; He Yu,
2005)
because imaging time of each line is very short, the correlation
of the exterior orientation elements of different lines is
inevitable, which might induce no solution for collinearity
equations. The ridge estimation has been used for solving this
problem, but it is difficult to choose ridge estimation parameters
in the classical ridge estimation algorithms. Zhang Yan et al.
(Zhang Yan et al., 2004) proposed a robust combined ridge with
shrunken estimator (RCRS), which could improve complex
collinearity of coefficient matrix and detect singularity of
observed value, making the estimator optimal and stable. Here
we used this algorithm to orient SPOT image and achieved
sub-pixel RMS error.
In (3) for airborne SAR image, j^ is approximately zero for
earth's rotation. But for spacebome SAR image, is not
approximately zero but a linear function with time since flight
eight is higher. Because y linearly varies with time, can
be expressed as a n -order polynomial with y . (Wang
Donghong et al., 2005; Chen Puhuai et al., 2001) Then (3) can
be written as
The range equation of slant range image is
(X - Xs ) 2 + (Y - Ys ) 2 + (Z - Zs) 2 = (y s M y + Ds 0 ) 2 ( 2 )
Where Ds is the slant range delay, y is the across-track
image coordinate of ground point P, M is the across-track
pixel size, ( x,Y,Z ) are the object space coordinates of ground
point P, ( Xs,Ys,Zs are 0 bject space coordinates of radar
antenna center.
Xv (X - Xs) + Yv(Y - Ys) + Zv (Z - Zs) ( 4 )
= a o + a \Y s + a iYs + a 3 yl + a *yt
Where «.(/' = 0,1,2,3,4) are polynomial coefficients.
When the exterior orientation elements of linear array
push-broom imagery and SAR image have been achieved
respectively, the stereoscopic pair is constructed. The space
coordinate of ground point can be computed with homologous
image points based on space intersection.
The Doppler equation is
Xv (X - Xs ) + Yv (7 - Ys ) + Zv (Z - Zs ) = - f DC ^
Where R is the slant range of ground point P, ^ is the
radar wavelength, j- is the Doppler frequency.
F. Leberl model is composed of formula (2) and (3). When
f DC = o, they can be linearized to achieve elements of exterior
orientation with gcps.
3. THE CONSTRUCTION OF COMPOSITE STEREO
MODEL
3.2 Space Intersection
A procedure in the composite stereo positioning is the
determination of object space coordinates X,Y,Z from a pair
of homologous image points ( x{ ,yf, x 2 >y 2 ) measured in both
images of a stereo model. Using these measurements as input to
the appropriate pair of mapping equations (i.e. equation (1) and
(2) (4)) yields 4 equations to calculate 3 unknown entities.
The overdetermined problem is solved by standard least squares
adjustment using equation (1) and (2) (4) to calculate proper
increments to the approximations given for the unknown
coordinates. Using equation (1) and (2) (4), the corresponding
position equation is as follows (in matrix notation):
V = DA - L P (5)
3.1 The Construction of Composite Stereoscopic Pair
There usually are two ways to construct stereoscopic pair. One
is first relative orientation with connection points and then
absolute orientation with gcps. The other is direct exterior
orientation of two images respectively and then combination of
them. Since the equations of relative orientation are derived
from collinearity equations, the former way is invalid for SAR
image. In the later way, the orientation procedure of two images
can be applied with two different groups of gcps respectively,
which avoids the difficulty for collecting connection points.
Here we choose the later way to construct the composite
stereoscopic pair.
The residual error vector is y - V2 Vj ^ y ;
The coefficient matrix is
D = (A D 2 D 3 D a ) t =
^d\\ d n d X2 f
d 2X d 22 d 2i
^31 ^23 ^33
V^41 ^24 ^43 J
The increments vector of the unknown ground coordinates is
A = {dX dY dZ) T \
As shown in (1), the exterior orientation elements of different
line in linear array push-broom imagery are different. But