C.O.G.
0.8,110.8,103.3)
11.3,64.5,104.1)
6.7,103.1,104.7)
5.0,137.6,104.7)
2.4,176.8,105.2)
7.1,164.9,105.1)
5.0,180.5,105.0)
0.8,198.7,105.0)
8.0,220.5,105.1)
33.8,217.1,105.1)
)9.3,179.0,105.3)
9.5,141.9,105.4)
2.1,101.7,103.6)
81.8,83.3,103.7)
49.7,79.7,104.4)
33.7,111.2,104.5)
218.0,9.1,101.4)
in a suburban area.
C.O.G.
3.8,217.1,105.1)
9.3,178.9,105.3)
) in the original tri-
detected houses in a
tom mesh).
URE WORK
ssion of huge sets of
ute force data com-
jecifically concerned
ires of the 3D scenes
nition. For this pur-
sh representation of
d extract important
preserved in coarser
lbeit at lower reso-
ng can be repeated
srving the same fea-
itationally attractive
t techniques. It sim-
f sensory range data
features can be used
nting the scene into
1996
nearly-planar patches. Meanwhile, we also define generic
models of 3D objects based on invariant relations between
sections of the surface boundary of our objects of inter-
est. These constructed models can be easily compared
to the coarsened meshes of the 3D range data. Applica-
tions of this approach range from interpretation of remote
sensing and photogrammetry data, digital terrain model-
ing (DTM), medical imaging, outdoor autonomous vehicle
navigation, planetary terrain mapping, geographical informa-
tion systems (GIS), CAD model reverse engineering.
In this context, we have identified several possible scenarios
of interaction between triangular mesh modeling and coars-
ening for the purpose of fast and accurate object recognition.
Although more effort is required to build a library of generic
object models such as the example we presented, the initial
results are very promising and indicate the generality and flex-
ibility of this approach. Future work will be focused on the
intelligent integration of visual intensity data with the avail-
able range data. We maintain that 3D object recognition is
best done using range information. Nonetheless, such addi-
tional information would greatly guide the object recognition
phases by clarifying possible ambiguities in interpreting range
data which tends to be more noisy than intensity data.
REFERENCES
(Boender et al., 1994) Boender, E., and Bronsvoort, W.F.,
and Post, F.H., 1994. Finite-element mesh generation
from constructive-solid-geometry models. Computer-Aided
Design, 26, pp. 379-392.
(Chen & Schmitt, 1994) Chen, X., and Schmitt, F., 1994.
Surface modelling of range data by constrained triangula-
tion. Computer-Aided Design, 26, pp. 632-645.
(Cheng et al., 1988) Cheng, K., Idesawa, M., and Soma, T.,
1988. Analysis and manipulation methods of geographic in-
formation. In: International Conference on Pattern Recog-
nition, Rome, Italy, pp. 897-900.
(Chew, 1993) Chew, L.P., 1993. Guaranteed-quality mesh
generation for curved surfaces. In: Ninth Annual Sympo-
sium on Computational Geometry, pp. 274-280.
International Archiv
(De Floriani & Puppo, 1988) De Floriani, L., and Puppo, E.,
1988. Constrained Delaunay triangulation for multiresolu-
tion surface description. In: Ninth International Conference
on Pattern Recognition, Rome, Italy, pp. 566-569.
(Fayek & Wong, 1994) Fayek, R.E., and Wong, A.K.C.,
1994. Triangular mesh model for natural terrain. In: SPIE
Conference on Intelligent Robots and Computer Vision
XIII: Algorithms and Computer Vision, Boston, MA, 2353,
pp. 86-95.
(Fayek & Wong, 1995) Fayek, R.E., and Wong, A.K.C.,
1995. Hierarchical model for natural terrain using topo-
graphic triangular meshes. In: Third IASTED International
Conference on Robotics and Manufacturing, Cancun, Mex-
ico, pp. 252-255.
(Fua & Sander, 1991) Fua, P., and Sander, P.T., 1991. From
points to surfaces. SPIE Geometric Methods in Computer
Vision, 1570, pp. 286-296.
(García, 1994) , García, M.A., 1994. Reconstruction of visual
surfaces from sparse data using parametric triangular ap-
proximants. In: IEEE International Conference on Image
Processing, Austin, TX, pp. 750-754.
(George, 1991) George, P.L., 1991. Automatic Mesh Genera-
tion: Application to Finite Element Methods. John Wiley
& Sons, Inc.
(Rippa, 1992) Rippa, S., 1992. Long and thin triangles can be
good for linear interpolation. SIAM Journal of Numerical
Analysis, pp. 257-270.
(Schroeder et al., 1992) Schroeder, W.J., Zarge, J.A., and
Lorensen, W.E., 1992. Decimation of triangle meshes.
Computer Graphics, 26, pp. 65-70.
(Suetens et al., 1992) Suetens, P., Fua, P., and Hanson,
A.J., 1992. Computational strategies for object recogni-
tion. ACM Computing Surveys, 24(1), pp. 5-61.
(Zheng & Harashima, 1994) , Zheng, W., and Harashima,
H., 1994. The automatic generation of 3D object model
from range image. In: IEEE International Conference on
Acoustics, Speech and Signal Processing (ICASSP-94).
pp. 517-520.
191
es of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
ERRORES
ET
