PHOTOGRAMMETRY AND COMPUTER GRAPHICS FOR VISUAL IMPACT ANALYSIS IN ARCHITECTURE
Peter Durisch
Informatik im Ingenieurwesen, ETH-Hoenggerberg
CH-8093 Zurich, Switzerland
e-mail: durisch@pfi.ethz.ch
ISPRS Commission V
ABSTRACT :
The paper presents a method to produce realistic computer-generated color images which aids to the judgement of aesthetic
properties of planned buildings. The planned buildings are embedded into the existing environment by photomontage. The pla-
nar universe of conventional photomontaging is extended to three dimensions. During an interactive preprocessing step, a
three-dimensional description of the existing environment is created: Geometrical data, atmospheric parameters and illumina-
tion parameters are retrieved from digital site photographs. The final image, combining the planned building and the environ-
ment represented by the site photographs, is rendered by an extended ray-tracing algorithm. The algorithm operates in three-
dimensional object space and handles the interaction between the building and its environment with respect to hiding, shadow-
ing and interreflection. For the simulation of illumination effects, material parameters are retrieved from the photographs at ren-
dering time.
KEY WORDS : Image Synthesis, Ray-tracing, Image Analysis, Feature Measurement, Visual Impact Analysis.
1. INTRODUCTION
Among the traditional methods which support the delicate job
of judging the visual impact of planned buildings on the land-
scape or cityscape are: images (plans, perspective drawings,
photomontages) and three-dimensional hand-crafted models.
Though there is a justification for all of these, observation re-
veals that they are either unable to show the subtle effects
which make up the overall appearance of a building, inflex-
ible for case studies, or expensive due to the invested manu-
al and artistic work. Thus, finding new methods to provide
accurate sources with which to judge the aesthetic properties
of planned buildings seems to be a worthy goal.
À fundamental requirement is that a building needs to be in-
tegrated into the existing environment (landscape, cityscape)
(Leonhardt, 1984). The paper presents a method to produce
realistic computer-generated color images which aids to the
judgement of aesthetic properties of planned buildings. The
planned buildings are embedded into the existing environ-
ment by photomontage. Photomontage is the combination of
artificial objects (the planned building) with a particular envi-
ronment represented by a number of digital site photographs.
During an interactive preprocessing step, a three-dimension-
al description of the environment is retrieved from the
photographs. In a second step, the photographs, the infor-
mation retrieved from the photographs and the CAD model of
the building are used to generate the final images. These
show the planned building embedded in the existing environ-
ment from the perspectives of the site photographs. See
also Durisch (Durisch, 1992).
2. PREVIOUS WORK
Computer-generated objects overlaid on background photo-
graphs are shown by Uno & Matsuka (Uno, 1979), Feibush
et al. (Feibush, 1980), (BDC, 1990), Takagi et al. (Takagi,
1990) and Doi et al. (Doi, 1991). Maver et al. (Maver, 1985)
calculate the camera's position and orientation in three-di-
mensional object space. The wireframe model of the planned
434
building is perspectively mapped onto the background
photograph and an artist completes the drawing within the
frame. Background objects hidden by the building are manu-
ally eliminated from the photograph. Then the corrected
photograph is transparently overlaid onto the image of the
building. Nakamae et al. (Nakamae, 1986) calculate the
camera's position and orientation in object space and re-
trieve atmospheric parameters and illumination parameters
from the photograph. The image of the building is generated
by considering the retrieved information and overlaid onto
the background photograph. Finally, it is covered by the man-
ually identified foreground. The approximation of the back-
ground scene by faces and polyhedra in object space allows
to generate shadows and to simulate various atmospheric
conditions. Object space coordinates are obtained from topo-
graphical maps. Terzopoulos & Witkin (Terzopoulos, 1988)
show an animation with deformable solid objects copied into
a real background scene which was approximated by planar
faces in object space.
Kaneda et al. (Kaneda, 1989) generate a three-dimensional
terrain model from cartographical data. The surface texture
of the model is derived from aerial photographs. Artificial ob-
jects are merged with the terrain by a method of Nakamae et
al. (Nakamae, 1989) for the combination of a foreground and
a background based on known depth information. Though
the three-dimensional terrain model allows any observer po-
sition, the approach is of limited use whenever details of the
environment, especially those visible in horizontal views, are
required. Thirion (Thirion, 1992) generates images for the
test of image processing algorithms. A three-dimensional
model of an existing building is retrieved photogrammetrically
from aerial photographs. The surface texture of the model is
derived from the same photographs. The three-dimensional
model allows any observer position if the geometry of the
scene is known with sufficient precision and if photographs
from suitable perspectives are available. Changes in illu-
mination are simulated by removing and adding shadows.
For this purpose, illumination parameters and material pa-
rameters are retrieved from the given photographs.