2.4 STRIP CONNECTION (S4)
A great number of gross errors, in phototriangulation, comes from
tie points. This step search for tie points errors. The detection
procedure can start when two strips have been formed, no matter if
the strips were obtained analytically or semi-analytically. The size
of these errors varies in a very large interval.
The algorithm is estabilished assuming that the form of the strips
are invariant. So they can be brought together through a movement
and scale. The functional model chosen is, then, the spatial
similarity transformation:
X X Xo
Y|= À R| y | + | Yo (2.4.1)
Z 2 Zo
where X, Y, Z represent a point, say, on the upper strip and x, jy,
z the same point on the neighbor strip down; R is the orthogonal
rotation matrix (R = Ë (uw, 6, k)) and A, à, $4, k, Xo, Yo, Zo are the
parameters of the transformation.
The tie points are used, in the adjustment, to estimate the
similarily transformation parameters. It is important a good
distribution of “tie ^ points, to :;allow . a. rigid geometric
transformation. The implicit adjustment model is applied, where the
tie points coordinates X, Y, Z, x, y, z are treated as observations,
and A, 0, $, k, Xo, Yo, Zo are the parameters. Then the residuals of
the observations are used to evaluate the quality and detect gross
errors in the points.
As it is known, the least squares adjustment scaters the gross
errors on the observations, making the detection and location
difficult and not clear. The robust stimation was used, in the
algorithm, to improve location of gross errors in this phase. The
weihgt, as sugested in the literature [08], [06], [11], is defined
in three stages: first P = I = identity matrix; second
P
EXP (-0,05(V/00) ^*^) (2.4.2)
and third
EXP (-0,05 (V/co) 2*9) (2.4.3)
rd
Il
where V stands for residual and Oo? is the "a posteriori" variance
factor.
In the proposed algorithm, weight was chosen:
- 448 -