] criterion
ommission
68 pages.
| industrial
grammetry
'equences.
3SN 0256-
in Civil
-383183-3.
jrammetric
y, practice
h Recent
Gomes et
N 90 5410
lobson, S.,
ment using
ric Record
neras and
nent, PhD
ylor, R.N.,
ge capture
leformation
Congress
(., Cooper,
Tracking of
ent, ISPRS
squences -
W1, pp.80-
sample
|
(6):
inalysis | of
s of Turtle
id Remote
'ariation of
C field.
ensing, 58
arke, T.A.,
Seeing the
ised digital
Yternational
( Sensing.
SSN 0256-
)5. Digital
> structural
lance and
, Australia.
practical
toring. 3rd
g Surveys,
}.
-covariance
Aanuscripta
Multi-Image Correlation for Digital Photogrammetric Measurement Systems
Heinz Rüther
Department of Surveying and Geodetic Engineering
University of Cape Town
South Africa
Graeme van der Vlugt!
IMETRIC SA
Switzerland
Commission V, WG II
KEYWORDS: Digital Photogrammetry System, Constrained Correlation, Multiple Images.
ABSTRACT:
A multi-image correlation (MIC) method for determining image correspondences for digital photogrammetric systems is described in
this paper. The MIC method was primarily developed as a robust algorithm for determining approximate parameters for the multi-photo
geometrically constrained matching (MPGC) method. The MIC algorithm uses the a priori information of interior orientation
parameters, distortion parameters and exterior orientation parameters of a multiple image set to geometrically constrain the correlation
search area. As this paper will show, the use of multiple oriented (and calibrated) images not only reduces the search space of the
correlation but also considerably strengthens the estimation of correspondences. Additional features of the method as well as its
reliability are also outlined here. The results clearly demonstrate that MIC is significantly more effective than traditional stereo-
correlation techniques (using epi-polar methods).
A typical photogrammetric application for this type of algorithm is the automatic measurement of surfaces. An outline of how the MIC
and matching process fits into a surface measurement system is described. Two surface measurements, using standard resolution CCD
cameras and the surface measurement software developed at the University of Cape Town, are reported and the results tabled.
1. INTRODUCTION
The efficiency of digital photogrammetric systems is mainly
dependant on their level of automation for determining point or
feature correspondences between the images. A multi-image
correlation (MIC) method for determining correspondences
between image patches/points in pre-oriented images is described
in this paper. The MIC method was developed as a robust
algorithm for determining approximate parameters for the well
known multi-photo geometrically constrained matching (MPGC)
method (Baltsavias, 1991). MPGC is a powerful least squares
method for accurately matching small image patches in multiple
image sets, however one of the principal difficulties in using this
technique is the determination of good parameter approximations
to ensure convergence of the least squares matching. The MIC
algorithm uses the a priori information of interior orientation
parameters, distortion parameters and exterior orientation
parameters of a multiple image set to geometrically constrain the
correlation search area. As will be shown, the use of multiple
oriented (and calibrated) images not only reduces the search
space of the correlation but also considerably strengthens the
estimation of correspondences. In addition, the MIC method
employs individual patch shaping in the various images to
estimate the effects of perspective differences caused by the
imaging geometry. This is useful for images which may be
rotated by, for example, ninety degrees. The patch shaping in the
MIC can then be successfully used as initial estimates for the
patch shaping employed by the MPGC method.
A typical photogrammetric application for this type of algorithm
is the automatic measurement of surfaces. Either natural texture
or projected light can provide a dense set of surface features to
measure. In the case of smooth surfaces, estimates of surface
shape from neighbouring regions can be used to automatically
compute approximations for the MPGC. However, to handle
surface discontinuities and irregular surfaces a more robust
method, such as the proposed MIC, needs to be employed. The
important features of the developed MIC algorithm and an outline
of how the MIC and matching process fits into a surface
measurement system, are described. Finally, two surface
measurements, using standard resolution CCD cameras and the
surface measurement software developed at the University of
Cape Town, are reported on and the results tabled.
2. HARDWARE
The hardware used for the development of the digital
photogrammetric surface measurement system comprised a low
resolution ITC CCD camera (795x596 sensor elements) with an
8mm lens connected to a Matrox PIP512B frame-grabbing card
installed inside a 486 DX-33 PC. The cameras were calibrated
using images of a point field of 101 retro reflective targets.
Typically about 15 convergent images including some with 90
degree rolls around their optical axis were acquired for each
calibration. Seven additional parameters modelling radial and
decentering lens distortion, pixel spacing uncertainty and image
shear (see Beyer, 1992) were always included as unknowns in a
! This work was performed while the author was at the University of Cape Town
499
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996
