TEM
SERIE
METHOD
Associate Professor University of Nairobi,
Department of Surveying and Photogrammetry,
4
P,ü,Box 30197, NAIRGRI,KENYA
INTERNATIONAL SOC
AND REMOTE S
ABSTRACT
frcuracy improvement is often a desired requi-
rement in many applications of photogrammetry.
An extension of this study into the topic of
variance reduction in data analyses hy use nf
multi-control variate method among other Monte-
Cario techniques is presented. Variance reduc-
tion and the consequent increase of accuracy of
the estimate is the motivation for the study.
Straight single control variate,regression-
adjusted control variate, concomitant control
variate are some other variations in common use.
The multiple-control variate method is particu-
larly applicable in photogrammetry as it invel-
vee pnoncon- trol object space coordinate para-
meters estimation using simultaneously a set nf
control object space coordinates.
The 3-dimensional coordinates nf a point can
be considered as functions of pholto-coordinates
in at least twe photographe, interior orienta-
tion parameters nf the camera and exterior B
rientation parameters nf photographe
used.Because of this functional relationship,the
coordinates of adjacent points are dependent.
Using this information in a judicipus way, it is
possible to improve accuracy of computed 3-D
coordinates of noncontroil object space points.
Control point coordinates could thefore be used
in variance reduction by multicontrol variable
method. Comparison of resulting accuracy esti-
mates would reveal efficiency and suitability
of the selected number and locations of points
for use as multiple control variates. Simpulation
would be the basis of data formulation and ana-
lyses.
KEY WORDS:
ariance reduction, Monte Carlo,
Close range, Si
imulation, Accuracy improvement.
y
n
a
1. INTRODUCTION
Monte Carle techniques are often used in many
srien- tific disciplines and a considerable part
nf this effort is in the area of variante reduc-
tion in systems analyses. The applicability of
some nf these techniques to the normal case of
terrestrial and close range photogrammetry has
been demonstrated in íNagaraja, 1990,91). Though
at first sight it might seem that sampling pro-
cedures only apply in case of simulation studie-
z,further reflection should indic- ate that
conceptually and practically, it should be pos-
sible to incorporate these ideas in reduction of
practical data. However, this paper deals with
a case study in simulation,
The subject of variance reduction has received,
outside photogrammetry,considerable attention
Y DF PHOTOGRAMMETRY
NG, CDMMISSI
DN V
and a number of methods have been developed.
Therefore, there are a few techniques that help
to increase accuracy and hence efficiency of
zimulations, sometimes substantially, by produ-
cing lest variable observations. Accordingly,
there is a need to study such a possibility.
Hillier and Liebermani1786) have given a tong
list of successful applications of simulation
studies.
simulation are nnt new in photo-
Applications of : ;
lications are still possik-
nrammetry but new app
le.Variance reduction techniques seek either an
increase in precisionidecreased variance) for a
fixed sample size or a decrease in the sample
size required to obtain a given degree of preci-
sion. Several authors have cautioned us in using
these techniques,I1f properly used, these techni-
ques can provide tresendous increase in the
efficiency of the model; however,if the intui-
tion is faulty and analyst does nat use a
reasonable design, the technique can also be
very unpredictable and actually increase varia-
nce tor some technigues.fecause nf this reason,a
systematic and thorough study of the selected
method is essential in adapting it for any spe-
tific application.
The control variables technique applies very
well when there is an equivalent to the process
we are simulating that can be treated theoreti-
cally, For example,in the normal case of close
range photogrammetry we have an eguivalence
between computation of coordinates of a noncon-
trol object space point and that nf a control
object space point, We can then simulate the Y-
coordinates {say} of a selected object space
point and that of the known control point simul-
taneously, using same random numbers in both
computations, Then the difference in the known
and computed coordinate of the control object
space point is an estimate of the correction to
be applied to the computed results of the selec-
ted object space point. This procedure,therefore
tuts put the variance due to common parameters
in the two processes leaving only the variante
nf the error of approximation. Obviously, this
should be of a lower order of magnitude. In
photogrammetry, as control point information is
generally available, it can be used as a controi
variate. This same ides can be further extended
te a multi-control variablgiMDCV) technique and
an attempt is made to present this extended
application study in this paper.
2. À SIMULATION EXPERIMENT BASED DN NORMAL CASE
in order to understand the capabilities of the
HCY technigue,it is necessary to set up a frame-
work for the simulation study. This aspect is
reported in this section. The assumed/true ob-
jectt space coordinates ite nearest mm), and the