OBJECT SPACE BASED CORRELATION WITH ADDITIONAL INFORMATION
Frank Boochs
Institut fir Angewandte Geodasie, Frankfurt/Main, Germany
Abstract:
The conception of a new algorithm will be presented, aimed at the determination of digital elevation models by means
of image correlation. Main features of the algorithm are the central managment of all image densities within the object
space, the parallelization of the point determinations for individually defined regions with homogeneous morphology,
the parametrization of the surface by functionals of greater complexity and the introduction and use of additional infor-
mations, used as support for the definition and computation of the surface parameters. Besides a detailed description
of the conceptional ideas, some first test result will be shown.
Zusammenfassung:
Es wird das Konzept für ein Verfahren vorgestellt, das auf die automatische Berechnung von digitalen Höhenmodellen
abzielt. Charakteristische Merkmale des Verfahrens sind die Übertragung der Bildinformationen in den Objektraum
zwecks vereifachter koordinierter Handhabung der verfügbaren Daten, die parallele Abwicklung der Punktbestimmun-
gen innerhalb an die Oberflächenmorpholgie angepaßter Regionen, die Einführung komplexer Parameterisierungen
des Oberflächenverlaufes und die Einbeziehung zusätzlicher Informationen, die zur Stabilisierung der Berechnungen
und zur Unterstützung der geometrischen Beschreibung der Oberflächen eingesetzt werden können. Neben einer
Beschreibung des Konzeptes werden erste Testergebnisse an natürlichem Bildmaterial vorgestellt.
Keywords: Image Correlation, Image Matching, DEM Generation, Surface Modelling
INTRODUCTION impression of the underlying geometrical and radiometri-
cal relations as long as they are typical for the object in
question. In connection with the knowledge about the
In the last few years photogrammetric procedures and geometrical and physical rules of the central perspective
instruments have shown a trend towards the use of it then is possible to verify the result of the localization
purely digital data including the images themselves. This guaranteeing high precision and robustness even in
gives the opportunity to transfer most of the photo- cases of complicate geometric and/or radiometric cir-
grammetric tasks to the computer, thus opening up new cumstances.
application fields and acquiring greater flexibility. One of For the human operator the performance of this task in
the most important aspects of this scenario for sure is general seems not to be critical, because of his know-
the reduction of time and costs by substitution of the ledge about the real world, his perceptual talent and his
human mensuration activities by equivalent computer experience with the mensuration process itself.
algorithms based on digital image data. The attempt to substitute this work by a computer
One of these mensuration domains is the determination procedure starts with the perception, meaning to locate
of object surfaces by profiling, contouring or discrete similar distributed densities within the images. The scale
point measurements within a stereo model. This task of algorithmic solutions for this similarity detection task
needs precise visual perception capabilities and extensi- ranges from correlation set ups to least squares algo-
ve, in most cases not explicitly documented interpretati- rithms of different complexity and feature based soluti-
ve work. ons. Each procedure has its advantages and disadvan-
The first task allows the localization of homologeous tages but all of them have the lack of beeing unable to
image points using the density variations within their solve the interpretation process. This restriction in
close vicinity and guarantees an adequate precision general is circumvented by the introduction of some
under the existing conditions (image quality, image simplifying model assumptions what in fact limits the
contrast etc.). , flexibility and partly dramatically reduces the perfor-
The interpretative process assures the recognition of the mance. Although these similarity detection algorithms
object or parts of it, supposed the density is of sufficient might still be subject to further improvements it seems to
significance. This information is very useful to get a raw be of interest to extend the model assumptions.
127
