STEREO MATCHING USING LINE SEGMENTS OF ZERO CROSSINGS
Anton Schenk:
Department of Geodetic Science and Surveying
Ohio State University
Columbus, OH 43210-1247, USA
Otto Hofmann
Messerschmitt-Boelkow-Blohm GmbH
Postfach 801149
8000 Munchen, FRG
Abstract
For more than twenty years, area correlation has been the preferred
method in photogrammetry to find conjugent points on digital images.
While in certain cases this method produces good results, in others, it
fails. The human visual System is remarkably adept at stereo matching and
is superior to area correlation. The theory of human vision is understood
as an information process, and computer models clearly indicate that the
method of stereo matching is not that of area Correlation. The paper
Summarizes the most important results, and a new method of matching
line segments of zero crossings is suggested to precisely determine a
sufficient number of points for the orientation of images, without the
need for approximate locations of those points.
Introduction
There is an increasing trend in photogrammetry to work with digital
imagery rather than with analog photographs, be it because some
applications may quickly require results (real-time photogrammetry), or
be it simply because of reasons for automating certain photogrammetric
procedures. Generating a digital elevation model is a typical example of
attempting to automate a task normally left to an operator. Of the many
processes involved in dealing with digital imagery, stereo matching or
the formation of the three-dimensional object space is the most challen-
- 602 -