a. Detec-
A. Grün,
c Extrac-
e Images,
D. Stall-
for Build-
al Report,
4 Zurich,
ion Using
Proc. of
, Austria,
id A.M.
on. John
AcKeown.
Multiple
op, Mon-
om Aerial
ogramme-
_eonardis.
arametric
Computer
ue, Czech
ule-Based
oofs. In
n images,
S. Stiles.
082.
ORTHO-MOSAICS AND DIGITAL ELEVATION MODELS FROM AIRBORNE VIDEO IMAGERY USING
PARALLEL GLOBAL OBJECT RECONSTRUCTION
Mikael Holm, George Denissoff, Kaj Juslin, Matti Paljakka, Markku Rantasuo, Susanna Rautakorpi
VTT Automation, Finland
E-mail: Mikael.Holm(Qvtt.fi
Commision Ill, Working Group 2
KEY WORDS: DEM/DTM, Orthoimage, Mosaic, Video, GPS, Global Matching, Object Reconstruction, Parallel Computing
ABSTRACT
One of the main obstacles for the use of video camera in airborne mapping applications is the small field of view compared
to the resolution of the video camera. Single video images cover only a small part of the area to be mapped. Therefore
automated methods to combine huge amounts of video images into single image-mosaics are essential.
In this paper a system under development is described, which will take as input thousands of aerial video images and will
output a digital ortho-mosaic and also a digital elevation model of the covered area. The "3D-image-mosaic" will be made
automatically using the methods of global matching or global object reconstruction. In this case an object based approach
is used in the matching. As this kind of methods are mostly very computation intensive parallel computation is used.
First, the ideas behind the system are described, including the use of GPS navigation data with the on-line digitization of
the images on the aircraft. Then, the present status of the development of the matching software is described, including
breakline detection and solving of the normal equations on a parallel computer.
1 INTRODUCTION
The effective use of airborne video imagery in mapping or
environmental monitoring tasks requires the use of auto-
matic methods to combine thousands of images into one or
a few image-mosaics. To make accurate mosaics a digital
elevation model (DEM) of the terrain is needed. As a DEM
of the area to be mapped in many cases does not exist, a
system creating the mosaics should be able to compute the
DEM itself, using the image data.
One algorithm capable of computing image mosaics and
DEM's from image data is global object based multi-image
matching, also called global object reconstruction. It is a
general model for digital photogrammetry, integrating area-
based multi-image matching, point determination, object
surface reconstruction and orthoimage generation. Using
this model the unknown quantities are estimated directly
from the pixel intensity values and from control information
in a nonlinear least squares adjustment. The unknown
quantities are the geometric and radiometric parameters
of the approximation of the object surface (e.g. the heights
of a DEM and the brightness values of each point on the
surface), and the orientation parameters of the images.
Any desired number of images, scanned in various spectral
bands, can be processed simultaneously.
This algorithm is a generalisation of the least squares match-
ing methods (Ebner et al., 1987, Ebner & Heipke 1988).
Similar concepts have been developed independently
(Wrobel 1987, Helava 1988, Weisensee 1992). A detailed
description of this matching algorithm and an evaluation
using synthetic aerial and real close range imagery can be
found in (Heipke 1990, Heipke 1992). The first controlled
tests of the approach using real aerial images can be found
in (Ebner et al., 1993, Holm 1994).
In this paper the present status of the ESPRIT-III GLORE
projective is described, where the object is to demonstrate
the capabilities of the method for the creation of massive mo-
saics of digitized aerial video imagery, and to create a pro-
totype system consisting of parallel hardware and software
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
for such usage. The work is funded by the Commission of
the European Communities.
The prototype system under development consists of two
parts. The first part is on the aircraft taking care of the dig-
itizing of the video images. The second part is on ground
taking care of the orthoimage mosaic and DEM generation.
The airborne part is described in section 2. In section 3 the
matching method is shortly described. In section 4 the im-
plementation on parallel hardware is described. The most
computation intensive part of the system is the solving of
the normal equations. The description of the methods used
can be found in section 5. In order to reduce the number
of unknowns and thus speed up the computation, irregular
DEM's are used. This is done using breakline detection as
described in section 6.
2 THE AIRBORNE DIGITIZATION OF VIDEO IMAGERY
The video images are captured and digitized onto the hard-
disk of a PC on board the aircraft during the flight. The
Super-VHS video camera is connected directly to the com-
puter. Traditional tape recorders are not used. The intervals
between the grabbing of the video frames, image brightness
etc. can be adjusted interactively using the PC and a piece
of software developed for this purpose. The grabbed frames
are stored as 24-bit truecolor images.
While the PC is digitizing video frames a laptop is gather-
ing navigation information using Realtime Differential Global
Positioning System (RDGPS) measurements. The laptop is
equipped with a DGPS card and Radio Data System (RDS)
receiver. The differential corrections for the GPS measure-
ments are carried out using the RDS coded correction infor-
mation sent by the national broadcasting company of Fin-
land. As a very cheap and simple system is used the ac-
curacy of the RDGPS coordinates — received once every
second — are in the order of 3 — 5 m. The laptop shows
the planned flight route and the real flight route graphically in
real time, as well as the metrical deviations from the planned
route. Using this equipment the pilot can easily keep on the
331
