ruc-
num
ion,
n of
hine
1pu-
MIT
sur-
don.
uta-
317:
visi-
‚kim
s of
. 98
lata.
À ca-
rep-
mo.
lysis
TOWARDS AUTOMATIC DTM VERIFICATION EXPLOITING
STEREO ORTHOPHOTOS
Michael Hahn
Institute for Photogrammetry — Stuttgart University
Keplerstrafe 11, D-7000 Stuttgart 1
Commission III
Abstract
Automatic procedures exist for the acquisition of Digital Ter-
rain Models (DTMs) from digital images as well as for the
computation of digital orthophotos. Both, DTM and or-
thophoto, are frequently derived within a common frame-
work, in which the human operator plays a minor role. The
situation is just inverse with regard to the verification pro-
cess. The usual way is to superimpose a wireframe or contour
line representation of the DTM onto the stereo images. Then
the verification is carried out step by step by detailed visual
control.
Quality control and verification of digital terrain models have
always been a problem. It is getting more demanding with
regard to the automatic generation of DTMs by digital im-
age matching, as also the verification should be automatic
as far as possible.
In this paper we outline possibilities of automatic DTM veri-
fication using stereo orthophotos. The automatic verification
is approached by exploiting the intensity differences between
the stereo partners. Three different ways are discussed based
on (1) regression analysis, (2) modelling radiometric differ-
ences by finite elements and (3) generation of orthophoto
pyramids. All three approaches result in a segmentation of
the DTM into usually small areas, which are not represented
completely by the model and the large remaining area which
is consistent with the imaged world. Experimental results of
simulations and real data indicate that by the computation-
ally more efficient approaches (1) and (3) the misrepresented
areas are located with higher significance than with the finite
element procedure (2).
Keywords: Image Analysis, DTM, Orthophoto, Stereo-
scopic, Data Quality, Verification, Image Interpretation
1 INTRODUCTION
Digital terrain models and orthophotos
Today digitization and digital processing of images are con-
sidered to be the basis for developing procedures which au-
tomatically produce standard products of photogrammetry.
The term “automation” seems to be inherently attached to
the technical development of each time period, even though
the quality of automation increases usually and often previ-
ously reached progress is incorporated in the actual one. In
the last decade remarkable progress is achieved especially in
the acquisition of digital terrain models (DTMs). Numerous
presented papers discuss procedures, in which the manual
233
measurement process is taken over from techniques of im-
age matching. In connection with the algorithms for surface
interpolation this leads to procedures in which the human
operator plays a minor rule. The reconstruction processes
presented recently, which solve the matching and interpo-
lation step within one framework, work on the iconic level
as well as the symbolic level of image description. State of
the art is that many of these methods are checked by de-
veloping prototypes. Moreover some of the prototypes are
implemented operationally and matured to productive sys-
tems (Ackermann and Krzystek, 1991).
A second standard product in photogrammetry are the or-
thophotos. In general there is a simple process of differential
rectification in which the (aerial) photo is reprojected to be
geometrically congruend to a map. In the case of digital im-
agery this geometric definition of a digital orthophoto still
holds. Using the terrain model and the orientation of the im-
age, the location of each picture element of the orthophoto
can easily be transformed to the aerial image. The intensity
value at this image point then can be found by resampling.
In some applications the area represented by an orthophoto
pixel is significantly larger than the area of an object repre-
sented by an image pixel, so that in this case strong smooth-
ing accompanies resampling. This means, that the whole
process is just a simple image processing procedure of differ-
ential resampling. If the area for which an orthophoto has
to be produced is not covered by one aerial image, pieces
have to be put together from the adjacent photos. Within
this process, called mosaiking, adjacent images are adjusted
radiometrically, for example, in order to smooth away linear
steps (Jansa and Guangping, 1990). That such an adjust-
ment is necessary results from a simple fact: the intensity
values in orthophotos of the same scene, resampled from dif-
ferent images (taken from different positions in the world) are
usually different. Assuming all geometric aspects (perspec-
tive projection, orientation, surface model) to be perfectly
known, these differences reflect noise (which we consider to
be a minor problem) and physical aspects of image forma-
tion. Unfortunately, the physical aspects of image formation
we are concerned with correspond to the difficult question:
what determines the intensity value at a particular point in
the image and consequently in the orthophoto ? There is
a whole chain of dependencies, which begins with the A/D
converter characteristics of the scanner or the digital camera.
Clearly how much energy arrives at a particular point in the
image depends on the energy emitted from the surface, on
the degree of absorption, on how the object is illuminated
and how it reflects light. In most cases is assumed, that the
reflectance model is a combination of a Lambertian model
and a specular model.
