You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Close-range imaging, long-range vision

P. Patias, V. Tsioukas, L. Sechidis
The Aristotle University of Thessaloniki, Greece, Dept. of Cadastre, Photogrammetry and Cartography,
Email : patias@topo.auth.gr
Commission V, WG V/1
KEY WORDS: Conjugate points, epipolar images, correlation, image matching, interlaced images
The user of a Digital Photogrammetric Station (DPS) usually extracts manually the conjugate points during the restitution procedure
using epipolar stereo-images. Although the manual procedure is the most secure way for this kind of job it is the most boring and
tiresome part of a photogrammetric project. One of the advantages of the digital era of Photogrammetry is the automation of the most
of such tasks. For example, many Digital Photogrammetric Stations have the ability of acquiring automatically conjugate points,
during the procedure of relative orientation of stereo pair images. The same algorithm can be applied during the restitution procedure
when epipolar stereo-images are used. The main algorithm that is used for this kind of job is the correlation between the patches of
the master (e.g. left) and the slave (e.g. right) images of the stereo pair. Additionally the least square image-matching algorithm can
be applied to enhance the results of the matched points and provide sub-pixel accuracy.
This method provides the best results under certain circumstances. The correlation procedure should provide a solution very close to
the correct one. Only under this condition, it is possible for the following image matching technique to provide the sub-pixel
accuracy of the matching points. When the correlation technique does not provide a good approximate solution (i.e. less than 2-3
pixel away from the correct conjugate point) it is unavoidable that the least square image technique will fail.
The proposed technique tries to fill the algorithmic gap between the correlation and the least squares image-matching technique in
the above-mentioned procedure. Actually it concerns a method that generally speaking resembles a digital filter. Two very important
advantages of the proposed method are speed and efficiency. It is much faster than the typical correlation and least square image
matching technique and can be applied directly to the interlaced stereo-image. By using the simple zooming technique on the
interlaced stereo-image it can provide sub-pixel accuracy of the conjugate point location. Furthermore, it can be used in automatic
restitution, by estimating the position of the conjugate points in real-time.
1. INTRODUCTION - PROBLEM STATEMENT provides accuracies of 0.1 pixels, but may also lead to errors in
the presence of high noise or when the correlation process does
The restitution process is considered to be the last part of a not provide a good initial solution.
digital photogrammetic project. It is certainly the most boring
and tiresome process and it contains the smallest amount of
automation in a DPS. Although, the conjugate point collection 2. IMAGE FOCUS-DEFOCUS - MATCHING
has been used for several process in a photogrammetric project
(image point matching is used during the relative orientation Something that has never been used before from the
and in the automatic procedure of DTM generation) it has not photogrammetrists in the past is the processing of the interlaced
been applied for automated restitution. image that supplies to the user the 3D view of the imaged
objects from their epipolar images, as one single image that is

This is due to several reasons and the most important is the
inability to determine accurately (less than one pixel) the
conjugate points on the images of a stereo-pair. On the
contrary, the manual collection of stereo-restitution is able to
do so. After all, manual editing of the automatic restituted
object should follow and since the editing procedure is much
more difficult than the original manual restitution the
automation is avoided.
Another problem that also appears is the great amount of time
that is demanded for this kind of job. In most cases the
combination of cross correlation and least squares matching is
applied. The first part of the automatic process (correlation)
supplies accuracies to the order of one pixel, while the second
process is a very time consuming algorithm, which optimally
characterized by some very specific features. This image can be
processed as any other image; digital filters may be applied on
it and image statistics may be used to provide useful
information. This image is actually two images in a composite
one, since its odd rows come from the left epipolar image and
the even from the right one. Its main feature is the extremely
high gradient along the y-axis that may appear when the image
does not provide a correct stereo viewing of the imaged object.
In the case when the roaming of the epipolar images brings into
coincidence the points of the epipolar images, providing a
correct stereo viewing, the gradient of the images even along
the y-axis is low. Concluding we may say that the high textured
images that are provided from the stereo viewing system are
not correct while low textured images are correctly matched.
is |
an (
in c
of t
the :
In n