il
he
La
he
of
1d
re
1n
ag
OBJECT RECONSTRUCTION USING 2-D PROJECTIVE TRANSFORMATIONS
Ilkka Niini
The Academy of Finland
Present: Helsinki University of Technology
Institute of Photogrammetry and Remote Sensing
SF-02150 Espoo, Finland
Abstract
A method for object reconstruction from arbitrary images is presented. Relative orientation of the images is based
on the 2-D projective transformations. As a result, the normal case of stereophotogrammetry is obtained. The
images are resampled along the rows in such a way that the new rows coincide with epipolar lines. Therefore
the digitization of the object is easily performed with matching along the rows. As an experimental example the
reconstruction of the Cathedral of Helsinki is presented, where the images are parts of digitized old viewgraphs.
KEY WORDS: Close-range, 3-D, Photogrammetry
1. INTRODUCTION
In conventional stereophotogrammetry object
reconstruction is usually performed using nonlinear
relative orientation, based on the coplanarity
condition. As a final result an object model is obtained
which can be digitized. Using digital images, this
procedure has been tried to be automatized, but these
algorithms have been more or less restricted to certain
applications.
In this paper an object reconstruction method is
presented which uses 2-D projective transformations
for relative orientation of images. These transform-
ations are linear in principle, and in addition, no
information about the interior orientation of the
images is needed. As a result of the relative
orientation, a stereopair of images is obtained which
satisfies the normal case of stereophotogrammetry.
After the orientation, the 3-D model coordinates of the
object can be produced along parallel epipolar lines.
The measurement stage of this method is a
combination of manual and computerized measure-
ments. In future, more automation is included. The
method is developed at the Helsinki University of
Technology.
The images may be anything between digital aerial
photographs and postcard views digitized with a grey-
scale scanner. Typically, digitized video images of size
512 x 512 pixels have been used.
2. RELATIVE ORIENTATION USING 2-D
PROJECTIVE TRANSFORMATIONS
The relative orientation method has its base on the 2-
D projective transformations instead of the perspective
collinearity condition. There is a more precise
mathematical presentation of the method in the paper
by Haggrén & Niini, 1990.
677
The method has three parts: 1) computation of
singular correlation between the images, 2)
computation of projective transformation coefficients
of both images from the correlation parameters, and
3) rectification of the digital images on a new plane,
which corresponds to the plane where the images are
asina normal case of stereophotogrammetry. Finally,
a 3-D object model can be computed using
conventional parallax equations.
For convenience, the main parts of the method are
presented here. Readers who are already familiar with
the mathematical concepts may continue reading from
chapter 3.
2.1 MATHEMATIC CONCEPTS
The camera model is assumed to be a pinhole camera,
ie. no radial distortion exists. The 2-D projective
transformation equations for the first image are
aux by;c
Xe 1514 Dy; C, (1a)
"o guy hy1
> > f
n dx eiy it 1 (1b)
^o gx hyÿ,+ 1
and for the second image
a9x";+ Daÿ";+ C;
n (2a)
xe gx'hjy'-1
E ur dx" e;y',- f, (2b)
'ogx'i+ hy'+ 1
where X’, Y;, and X", Y", are the new image
coordinates of the first and second image at any
arbitrary plane. The x,y’; and x",,y", are the original
image coordinates. Index i is the number of the point
(i=1...n).