ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002
3. METHODOLOGY
3.1 Optimum Sequence for Parameter Estimation and
Initial Correspondence Determination
To execute the MIHT for SPR, we must first make a decision
regarding the optimum sequence for parameter estimation that
guarantees quick and robust convergence to the correct solution.
Various image regions are affected differently by changes in the
associated EOP. Some parameters have low influence on some
regions while having larger influence on others. Therefore, a
certain region in the image space would be useful for estimating
some parameters if they have a large influence at that region
while other parameters have minor or almost no influence at the
same region. Moreover, optimum sequence should not affect
previously considered — regions/parameters. Conceptually,
optimal sequential parameter estimation should follow the same
rules of empirical relative orientation on analogue plotters
(Slama, 1980). The following paragraph deals with how to
determine the optimum sequence for parameter estimation
together with the corresponding regions for their estimation.
rd
1 3
X
4 5 6 11
7 8 9
Figure 1: Image partitioning.
For such objective, we have divided the image into nine regions
labelled from 1 to 9 as shown in Figure 1. Regions 2, 5 and 8
have small x coordinate values (ie. x; € x, = xg = 0), while
regions 4,5 and 6 have small y coordinate values (i.e. y, z ys —
ye * 0). The collinearity equations (Equations 1) have been
linearized and reduced by assuming small rotation angles,
which is the case of vertical aerial photographs (Equations 2).
2
C X xy X
e. «—dX,-t—dZ,———doe-4(c-—)dó-- vdrK
IH: S Hic OT (2)
2
gs di d dz cn dos dd xdi
rH H C €
In Equations 2, the terms e, and e, represent image space
displacements in the x and y directions resulting from
incremental changes in the EOP (dX,, dY,, dZ,, dm, dà, dx). It
has to be mentioned that Equations 2 are not used for the
parameter estimation. Instead, we will use them to identify the
influence of the EOP on various regions in the image (Figure
1). Table 1 summarises the effect of incremental changes in the
EOP on the nine image regions (Figure 1).
By analysing Table 1 and following the previously mentioned
rules at the beginning of this section, the optimum sequence for
parameter estimation is as follows:
1. Use points in region 5, to estimate X, and Yo.
2. Use x-equations of points in regions 2 and 8, and y-
equations of points in regions 4 and 6, to estimate X:
3. Use x-equations of points in regions 4 and 6, and y-
equations of points in regions 2 and 8, to estimate Z,.
4. Use points in regions 1, 3, 7 and 9 to estimate œ and @.
This sequence will be repeated, after updating the initial values
for the parameters with the estimated ones. The procedure can
be described in the following steps.
Sweep 1:
e Establish approximations for Zo, c, Gand x
eo Determine the range and the cell size of the accumulator
array for (Xy Yo) depending on the quality of the
approximations of the other parameters.
* Using the collinearity model, solve Xy, Yo for every
combination of object point with one image point in region
S.
e At the location of each solution,
corresponding cell of the accumulator array.
e After considering all possible combinations, locate the
peak or maximum cell of the accumulator array. That cell
has the most likely values of X, and Y,.
increment the
Sweep 2:
Repeat sweep #1 for (X) (Zo) and (c, 4$) updating the
approximations of the parameters, while using the appropriate
regions that was determined earlier.
Sweep 3:
Decrease the cell size of the accumulator arrays for (X,, Yo), (4),
(Zo) and (c, ¢) to reflect the improvement in the quality of the
approximate EOP. Then, repeat sweeps 1-3 until the parameters
converge to the desired precision.
Table 1: The influence of different image regions on the
arameters.
Region dX, dY, dZo
X eq. y eq. x eg. y eq. x eq. y eq.
1 c/dZ 0 0 c/dZ -x/dZ | y/dZ
2 c/dZ 0 0 c/dZ 0 y/dZ
3 c/dZ 0 0 c/dZ x/dZ v/dZ
4 c/dZ 0 0 cdZ | -x/dZ 0
5 c/dZ 0 0 c/dZ 0 0
6 c/dZ 0 0 c/dZ x/dZ 0
7 c/dZ 0 0 c/dZ -x/dZ | -y/dZ
8 c/dZ 0 0 c/dZ 0 -y/dZ
9 c/dZ 0 0 c/dZ x/dZ | -y/dZ
Region do do dx
x eq. y ed. X eg. y eq. x eq. y eq.
1 xyle. | -c-/c | ete -Xy/c y x
2 0 -c-y,/c C 0 y 0
3 -xyle | cle | c*x'le |. xylc y -X
4 0 -C C+Hx“/c 0 0 X
5 0 -C c 0 0 0
6 0 c lem. 0 x
7 -Xyle | -c-y)le | e+xle xylc -y x
8 0 -c-y)/c C 0 -y 0
9 xylc. | -c-y le | cx lc -xy/c -y -X
One has to note that the lack of features in any of the nine
regions may only slow the process. The reason is that all EOP
affect all regions but with different magnitudes. Only the
maximum influences/contributions are represented in Tablel.
A - 152