International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Case | Case 2 Case 3 Case 4
Sq] Ell.].Sg. l.El- 1.Sq. 1 El. | Sq. 1 Ell
Templat
e sizes
used
(the last
one is |l13x13
the one |29x29
which
returns
the
point)
29x29
13x13|25x25|25x25137x37129x29| 15x 15115x15
41x41
Pixel
decreas
e - 1.2% - 1.4% - 0.6% - 0.9%
percent
age
Iteratio
ns on
the last
templat
e used
[o»]
La
oo
oo
1
LI
(2
t2
N
=
2
I
to
Go 25.14 2:12]135231. - 7.81 | 10.54 | 7.
Returne
; XES YES | YES LYES.| NO | YES |-YES | YES
d point
Correct
NO | YES | NO | YES | NO | YES | YES | YES
match
Table 1. Comparison of results.
The only problem that arises from the application of the
proposed method is the complication of the calculations, but
then again this is the only way to attain better results.
Complication of calculations leads to more computer time.
The algorithm has not been timed since it has been used until
now manually and time differences cannot be observed, but it
is predicted that it will be slower than the square.
It should be mentioned though that not much attention should
be given in speed, because computer power doubles every 1.5
years. When LSQM was first introduced it was very slow for
the contemporary computers, not to mention the quality of
CCD sensors. Today, matching over a whole model,
producing 18000 points can be completed in 3 minutes and
for 2,5 million points in 30 minutes in an average computer.
The problem might be evident when applied in DEM
collection. Prior to this code optimization will decrease the
algorithm’s speed by half. Use of fewer vertices to describe
the ellipse is another possible source of time saving.
Reduction by 20%, will save time almost 15% over the whole
matching algorithm. Another interesting feature is that after
the second iteration the ellipse does not change considerably
both is shape and orientation, meaning that it is almost
useless to reform it after each iteration. This is another point
where processing time can be saved.
In 2 cases (1 and 3), the square method fails in the suggested
template and uses a bigger one in order to find a solution. In
the same cases ellipse uses the suggested template and is
equal or faster, in terms of total iterations, in all cases.
Therefore the ellipse method compensates speed, up to a
point by itself.
Epipolar geometry provides a very good solution for linear
features nearly perpendicular to the epipolar line (Baltsavias,
E., 1993). The ellipse provides solution for all linear features,
without the use of relative orientation, and therefore is a
universal method, while being more accurate for all points.
Failure rate, including points returned from square template
as correct, without being so, is reduced considerably. Hence
it is safe to conclude that it is a promising method which
requires further research. The next step is speed optimisation
and application on DEM collection of different objects and
scales to verity these conclusions. Comparison will be done
against square template DEM as well as against a reference
DEM.
References
Forstner, W., 1986, On feature based correspondence
algorithm for image matching and least squares matching,
International Archives of Photogrammetry and Remote
Sensing, Vol. 26, P3, Rovaniemi, pp. 150-166.
Gruen, A.W., 1985, Adaptive least squares correlation, Sourh
African Journal of Photogrammetry, Remote Sensing and
Cartography, 14(3):175-187.
Gruen, A., Baltsavias, E., 1985, Adaptive least squares
correlation with geometrical constraints, Proceedings Of
SPIE (The Society of photo-Optical | Instrumentation
Engineers), Vol. 595, pp. 72-82.
Rosenholm, D., 1986, Accuracy improvement of digital
matching for evaluation of digital terrain models,
International Archives of Photogrammetry and Remote
Sensing, Com HII, Vol. 16 —3/2, pp. 573-587.
Rosenholm, D., 1987, Empirical investigation of optimal
window size using the least squares image matching method,
Photogrammetria, Vol. 42, pp. 113-125.
Baltsavias, E., 1991. Multiphoto Geometrically Constrained
Matching. PhD thesis in Institute for geodesy and
photogrammetry, Zurich, 221 p.
Skarlatos, D., 2000, Image matching towards maturity,
International Archives of Photogrammetry and Remote
Sensing, Amsterdam 2000, Vol. 33, Part B3, pp. 845-849.
Balodimou, A., M., 2000. Theory of errors and position
fixing, ambiguities in 2 and 3 dimensions. Notes (in greek)
from lectures in National Technical University of Athens. 24
p.
Atkinson, K., B., 1996. Close range photogrammetry and
machine vision. Whittles Publishing, BristolScotland UK,
371p.
Guelch, E., 1988, Results of test on image matching of
ISPRS WG I11/4. International Archives of Photogrammetry
and Remote Sensing, Kyoto 1988, Vol. 27, Part B4, pp. 254-
271.
Trinder, J., C., Becek, K., Donnelly, B., E., 1992, Precision
of image matching, International Archives of
Photogrammetry and Remote Sensing, Washington DC, Vol.
29, Part B3, pp 820,823
Baltsavias, E.P., Stallmann, D., 1993. SPOT stereo matching
for digital terrain model generation. Proceedings of 2" Swiss
Symposium on Pattern Recognition and Computer Vision,
Zurich, pp. 61-72.
SA
all
