lem concerns to
tive positioning,
lation of relative
n use absolute
terrain or the
)0]) for metric
rivileged role as
n.
form a sweep of
se a sequence of
as. "There ‘exist
quadrilaterals in
jn, as the frames
ways ([Har00]).
ompare different
yut parallel lines.
altiple (triple or
ng quadrilaterals
n inside). So, we
] by perspective
compare (and
sive images. To
quadrilateral we
B-- DC) ^ AC
ween homologue
of homologue
ry ([Har00]). The
imposes strong
elements along a
ansformation is
nslation vector
re of gravity of
from a standard
econd of latitude
on of parameters
ctors. Hence, we
with a common
ue quadrilaterals
is which give an
etween sampled
ferences between
e quadrilaterals
estimator of the
sed quadrilateral
map. Trapezoids
two orthogonal
nal view. So, we
s parallel to the
ved by discarding
e of a segment
tive optimisation
mulation and the
thod is based on
energy functional
ion of L'- and L’-
1al potential and
kinetic energies. In our case, we incorporate an additional
elastic term to the energy functional which provides a more
flexible tool for Digital Elevation Maps in the 3D Case.
i
; ; AL di ;
n - p^y(a *)pe! -av1]-2p^ Y. ET +22(1+ 5%).
i 1 j
2
where 4; and 4, are the intensity values of two pixels, p is the
correlation between those two pixels and à, b, and b, are
parameters of both images. 3) the estimation of the disparity
gradient to evaluate Lagrangian optimal conditions, and 4) the
extension of usual superimposed Delaunay triangulations to
give solutions compatible with elastic models. The quadratic
part is the responsible for the data adjustment, the linear term
penalties meshes with a large number of vertices and the elastic
term is in charge of an adaptive behaviour towards a Pareto
optimal solution. In our case, solutions of this multiobjective
optimisation adjust in an adaptive way to the boundary of a
quadrangular region. So, we extend the ordinary classical
approach based on ordinary L'- and L’-norms. Optimal
solutions for the classical case are given by Delaunay
triangulations which maximize the minimal angle ([Ede91]). In
our case, we can not expect an optimal result as good as above,
because multiobjective optimisation have no necessarily a
unique solution, but only Pareto optimal solutions with
smoothness and adaptive properties.
6. Conclusions and future work
We have developed grouping and tracking methods based on
quadrilaterals. This approach provides tools which are more
naturally linked to the contours structure contained in aerial
images. So, we avoid some ambiguity problems linked with
more traditional triangulations. We introduce a functional for
multiobjective optimisation, and we reinterpret some properties
of solutions in terms of superimposed structures. The next step
to be given is to evaluate the invariance in terms of histograms
of frequencies linked to the same region seen under different
orientations and under different lightening or atmospheric
conditions. Instead of looking at the invariance in terms of
rotationally invariant Hamiltonian fields (as in the Euclidean
case) we must look at invariance in terms of vector fields
preserved by the affine group. The robustness and accuracy of
this method is the challenge for the next future.
Hence, some specific ingredients of our approach to the
Multiobjective Optimisation are based on 1) the construction of
a disparity map along pairs of consecutive images, 2) the
correlation estimation, considering the maximum likelihood
estimator of p ([Ybe01 ]):
A, A; T 2p A, A;
2 =0,
i b,b, Llp b,b,
7. References
* [Ber97] M.de Berg, M.Van Kreveld, M.Overmars and
O.Schwarzkopf: “Computational Geometry. Algorithms
and Applications”, Springer-Verlag, 1997.
e [DeP02] I.De Paz: “Mathematical Models for GIS”, M.Sc.
Thesis, Univ of Valladolid, June 2002.
eo [Ede91] H.Edelsbrunner and T.S.Tan: “4 quadratic time
algorithm for the minmax length triangulation”, Proc. 32™
IEEE Symp. Found of Computer Science, IEEE Press,
1991, 414-423.
e [Fin02] J.Finat, M.Gonzalo-Tasis: “Dynamic trapezoidal
maps for coarse perspective models in indoor scenes”
Proceedings ISPRS 2002, Corfu, Greece, September 2002..
e [Har00] R.Hartley and A.Zisserman: “Multiple View
Geometry”, Cambridge Univ. Press, 2000.
e [Kre97] M. van Kreveld, J.Nievergelt, T.Roos and
P.Widmayer, eds: “Algorithmic Foundations of GIS”,
Springer-Verlag, 1997.
e [Rus99] J.C.Russ: “The Image Processing Handbook (3™
ed)”, CRC Press, 1999.
e [Vil02] A.Viloria, J.Finat and M.Gonzalo-Tasis: “A fast
self-organized iconic segmentation and grouping based in
color", Proceedings ISPRS 2002, Corfü, Greece,
September 2002.
® [Ybe01] Y.Belgued et.al.: “An accurate radargrammetric
chain for DEM Generation", Alcatel Space Industries,
ESA-ESTEC Publications, 2001.
—499—