001. A self
| optical flow
27-433.
. Real-time
ving vehicle.
John Wiley
nage analysis
ystem. IEEE
stems, 1(2),
on technique
background
Recognition
comparative
v estimation.
and Remote
[II, Part B5,
> for vehicle
tforms. In:
10te Sensing,
B5, pp.277-
. W*: Real-
es. IEEE
Intelligence,
h to moving
isactions on
p.577-589.
11, M., 2000.
ntersections.
"stems, 1(2),
ind. Nevatia,
deo stream.
1 Machine
olor Image
enter; visual
lications of
w approach
Journal of
CONSTRUCTION OF CITY MODEL BY MOTION
Yousuke Kashimoto *, Heitoh Zen”
“ Tokyo University of Mercantile Marine, Japan
kyousuke@ipc.tosho-u.ac.jp
? Chiba University, Japan
zen@imit.chiba-u.ac.jp
Commission VI, WG VI/5
KEY WORDS: Motion Stereo, Epipolar Plane Image Analysis, Digital Map, DP Matching , Texture data
ABSTRACT:
It is useful to construct the urban city model in three dimensional space in the various fields such as the navigation, walk-through
systems, and so on. We have researched a method to construct the city model based on the motion stereo with the image sequences
taken from a camera on the vehicle in addition to use an existing digital map.In this paper, we propose a method to generate the city
model with the information about its height because the tall building of our previous city model had no texture on the upper part and
its height was invalid. The height of a building is given by the result of the analysis about an aerial photograph. Then, we make the
texture on its upper part by devising the EPI (Epipolar Plane Image) analysis which is one of the motion stereo techniques.
1. INTRODUCTION
The 3D model of urban scene is widely used in various fields,
such as city engineering, traffic engineering and amusement.
Under present conditions, these models are made manually with
CG modeling tools. Hence, huge time and cost are wasted, and
application of these models is restricted. Therefore, in order to
construct model efficiently and automatically, many approaches
have been suggested. These approaches are roughly divided into
two methods. One is expression using stored 360deg. real
images from observation point without The 3D data[1][2]. This
method is called IBR (Image Based Rendering), it is possible to
realize photorealistic urban scene. But on the other hand, it is
necessary to use a large quantity of data to utilize many images.
The other is to make model using measured the 3D date of
buildings with the stereo method or a laser measurement[3].
This method enables to move a viewpoint freely and construct
model with a small quantity of data. We have constructed urban
scene model based on the 3D data too. We proposed a method
of modeling using measured the 3D data of buildings and
texture data of target buildings with motion images taken from a
camera on a vehicle[4]. The 3D data were obtained by applying
Extended EPI (Epipolar Plane Image) analysis that had been
improved over EPI analysis proposed by Bolles et al[5]. And
the texture data were obtained from the slit spatiotemporal plane
image which exchanged from image sequence. We express the
city model by doing matching of the 3D data and an existing
digital map. However, because it was not got height information
of a building using conventional technique, we have to give
uniform height but not individual height for each building.
When we consider our model as a practical city model , it
becomes necessary bird indication looking at a model from the
sky using the model. To resolve the above problem we utilize an
aerial photo analysis result. The texture of the upper part of a
building was not able to be gained in creation of the texture
from the slit image of the conventional technique. The method
using a fish-eye lens as a method of acquiring the texture of the
upper part of a building and the method using an Omni camera
can be considered. The texture of the upper part of a building is
gained by projection conversion of an image. Therefore, in this
paper, we give height information using an aerial photo analysis
result in addition to construction technique of the city model
that utilized three dimensions of information and a digital map
of a building by traditional image analysis and suggest
acquisition technique of upper part texture of the tall building.
First, we mention the method how to obtain 3D data of
buildings in Section 2, secondly Section 3 mentions a method to
utilize digital map data and aerial photograph analysis result, in
the third, Section 4 mentions a making method of texture by
projection conversion. Experimental results are shown in
Section 5, and Section 6 describes conclusion of this research.
2. IMAGE SEQUENCE ANALYSIS
2.1 Obtaining depth data by EPI analysis
3D data of buildings are obtained by EPI analysis. The traces of
feature points appear on the Epipolar plane image when the
camera motion is uniform linear motion (see Figure 2). These
traces are straight lines if the optical axis of a camera is
perpendicular to the camera path. The distance from the camera
path to the target buildings is able to be measured the slants of
these lines.
It is necessary to prepare many cameras for wide area
observation in an EPI analysis. In an EEPI analysis, all flows of
stationary objects appeared from FOE (Focus of Expansion) are
picked up by the camera whose the optical axis is parallel to the
camera path (see Figure 3). Thus a wide region can be observed
Figure 1: Trace of feature point on EPI
—567-
