1385
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
3.1 Ground Control Points
GPS-VRS is used to collect ground control points. The
selection of GCP was carefully observed based on distinct
features of ground to image such as road cross-sections and
centre of bridges. Even 22 points are observed scattering in the
whole image; 14 points are selected as GCP and other 8 points
are defined as verification check points. All control points
(GCP and CP) are located visually observation on the image.
3.2 Verification of Transformation Models
In general photogrammetry, a model defines a set of interior
orientation parameters (Wolf, 1983; Mikhail et al., 2001). and
exterior orientation parameters In this study, the transformation
of ground objects (x,y,z) to image coordinates (w,v) could be
done by 3D projective function using parameters of vendor
provided replacement sensor model (RPC), when the physical
sensor model is restricted to public. The coefficients of RPC
model ca be variable based on the sensors. Without knowing
the sensor information RPC coefficients are using in
photogrammetric processing such as ortho-rectification, DSM
generation without noticed accuracy loss (Grodecki, 2001).
Firstly, vendor provided RPC model is observed. The residual
errors of the model result are showed in the table 1. Then, the
errors of RPC coefficients are verified by the coefficients of
accurate GCP. As of transformation model, the 3D Projective
function is used. The function some time called as direct linear
transformation (DLT) model can be represented by equation 1.
u _ (a x x+a 2 y+a 3 z+a 4 )
(a 9 x+a l0 y+a u z+l)
v _ (a 5 x+a 6 y+07Z+a 8 )
{a 9 x+a x() y+a n z+\)
(1)
where a¡, a 2 , .... an are unknown linear orientation parameters
between two dimension image space (w,v) and three
dimensional objects space (x,y,z). The result of the model is
showed in the table 2.
4. THREE DIMENSIONAL MEASUREMENT
The photogrametric model can be easily understood using
simple below figure (2).
perspective center O
Figure 2: Simple photogrammetric model.
To extract 3D information from satellite images, at least two
stereo images, interior geometry of the sensor and exterior
orientation parameters is needed.
The study selected backward and nadir looks of ALOS-PRISM
as stereo pair (figure 3). Moreover, the coefficient parameters
are calculated with DLT model using GCP.
DSM extraction from triplet image
Forward y Nadir v Backward
(«bM)
ti = f(x,y,z)
OOO ps \
v = f(x,y,z)
OOO
OOO
Figure 3: DSM extraction from PRISM data.
4.1 Stereo Image Matching
Fundamentally, satellite orbit and attitude information are used
to generate DSM. Nadir-forward and nadir-backward pairs from
triplet images are used to calculate parallax with tie-points by
image-matching technique (Tadono, 2006). To generate tie-
points automatically, least squares image matching is a
powerful method to extract correspondence pixels from stereo
pair (Gruen, 2005). In the method of least squares, the
geometric differences can be modeled by affine transformation,
while the radiometric differences are modeled by an additive
and a multiplicative parameter. The main purpose of the
method is to refine the approximate matching positions and to
get high accuracy. The basic algorithm can be formulated in
three dimensions with the following generalized equations:
g,(x„y„z l ) + n(x l ,y l ,z,) = g s (x°, dx, y° s + dy, z° s + dz) ( 2 )
A g + n = ^k dx + ifdy + iirdz) ( 3 )
where g, (x,y,z), g s (x,y,z) are grey level function of the target
and the search windows; n(x t ,y,,zJ is error vector;
dg s dg s dg s
dx ’ dy 9 dz are gradients in x, y and z directions, which
can be declared as g x , g y , g z .
If both the radiometric quality and the geometrical differences
are considered, the equation (2, 3) can be written as:
g, (*,,y,, z ,) = K + h, * g 5 [(a 0 + a x x + a 2 y + a 3 z),
(b 0 + è,x + b 2 y + b 3 z),(c 0 + c l x + c 2 y + c i z)]
(4)