FACETS STEREO VISION (FAST VISION) APPLIED TO DIGITAL COLOUR IMAGES
B. Kaiser, B. P. Wrobel
Institute for Photogrammetry and Cartography
University of Technology Darmstadt (Germany)
International Society of Photogrammetry and Remote Sensing
Commission III, WG III/2
XVIII ISPRS Congress Vienna, Austria, 9-19 July 1996
KEYWORDS: Photogrammetry, Matching, Reconstruction, DEM/DTM, Orthoimage, Colour, Experiment
ABSTRACT:
In this paper a modification of the well-known object space based Facets Stereo Vision (FAST Vision) method is introduced.
It works with vectors as observations in each of the pictures being processed for surface reconstruction. Former versions of
FAST Vision only used scalars for each pixel as input values. The new version makes it possible to exploit the full
information of multi-channel imagery. The formal modifications of the method are presented in this paper as well as an
example of a surface reconstruction applying the new method to a pair of colour aerial pictures.
ZUSAMMENFASSUNG:
In dieser Arbeit wird eine Modifikation der wohlbekannten objektraumbasierten Methode des Facetten-Stereosehens
präsentiert. Sie arbeitet mit Vektoren als Beobachtungen in jedem der Bilder, das für die Oberflüchenrekonstruktion
herangezogen wird.. Bisherige Versionen von FAST Vision benutzen Skalare als Eingabewerte. Die neue Version macht die
volle Ausnutzung von mehrkanaligem Bildmaterial móglich. In dieser Arbeit werden sowohl die formalen Modifikationen
der Methode als auch ein Beispiel einer Oberfláchenrekonstruktion aus farbigen Luftbildern beschrieben.
1.Introduction
The major difference concerning the exploitation of data
between digital and analytical photogrammetry consists
in the use of colour information. Whereas the input for
analytical photogrammetry usually consists of colour
images, the input for digital photogrammetry are usually
pictures with 256 (8 bit) grey values. This results in a
considerable disadvantage for the digital method. A
simple visual comparison of grey value and colour
images depicting the same scene illustrates how much
information is lost in grey value images. The same loss of
information occurs when other channels of a picture, e.g.
a infra-red channel is not used as input for surface
reconstruction. The reason for the limitation of digital
photogrammetry to grey value images consisted in the
amount of image data which increases threefold in the
case of three colour channels. But nowadays, the price of
Storage media with large capacity can be neglected. As
the data not only has to be stored but also processed,
colour images require an improved processing
performance of the computer. This performance has not
ceased to increase with every new generation of
computer processors in recent years and won't do so in
coming years. Therefore the complete exploitation of
image information in digital photogrammetry becomes a
possibility which should be realised (Weisensee, Wrobel,
1991).
2.Facets Stereo Vision
The method for object reconstruction discussed in this
paper is an object space base one: Facet Stereo Vision
960
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
(Wrobel, 1987). It is based on inverting the process of
image formation using a finite element model of the
object surface and of the object grey values consisting of
patches called facets. Generally, functions T°, T"'..
describe the transformation of image grey values
G(x.y LG 'xX¥) at pixel positions (x,y),
(x,y) to object grey values G(X,Y) at a position (X,Y)
in object space.
T'(G'G',y)J- GQCGY)
(1)
T"{G"(x",y") }= G(X, Y)
If outer and inner orientation and approximate values
ZEN for the object surface are known the grey
values in the vicinity of (X5, v^ can be described by a
Taylor series.
G' (X9 dx, Y" *-dY4-aG* (dX, dY)- G'(X*, Y?)
ec (X5. eG (X^, Y?)
V ay
dY-- dG' (X, Y).
(2)
Inserting (2) into (1) yields:
Gy) T: (G'CX*, Y*)4
80 T) qn dG* (X, Y)).
The image rays passing the point of projection
(X 5, Y'5,Z'y) yield a formula linking differences dX and
dY with differences dZ:
CCE, |
ox 3)
