International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
matrix AB of the delta values. While a straightforward method,
careful attention to level of error is very important. Thus, we
can decide the spatio-temporal model accuracy criterion:
- Spatial accuracy: depending on the spatial geometric
errors, which include the plane errors (X, Y) and
elevation error Z, with the unit of the meter.
- Temporal accuracy: depending on the projection error
on the time axis, which can be substituted for time
matching ratio, with the unit of percent (9o).
- Attribute accuracy: depending on the agreement
percent sith the practical attributes, with the unit of
percent (%).
4. SPATIO-TEMPORAL MODELLING METHOD
FROM AIRBORNE SAR IMAGES
4.1 Required materials
To reconstruct the precise spatio-temporal model from airborne
SAR imagery, the required materials is must be considered
from space and time aspects.
On the one hand, it is necessary that the multi-temporal
airborne SAR images are available, with the time resolution of
the day. The airborne SAR images of the same rescarch region
should be acquired with the time interval of one day.
And the other hand, we need the stereo images of the same
region and time to reconstruct the 3D spatial model. And, GIS
basic graphics can provide us with the ground objects attributes
and land use information.
4.2 Data processing
For reconstructing the spato-temporal model cell, the object-
oriented spatial reconstruction model cell with special time
tamp must be considered. At the present, there has been many
methods to extract the Digital Elevation Model (Herland, 1997;
Yang, 2001; Peng, 2000) and 3D reconstruction spatial model
from SAR images (Tsay, 1994).
Firstly, the stereo airborne SAR images can be processed to
extract DEM of the model cell, and acquire the high-resolution
digital orthoimages.
Then, utilizing the imagery segment and height estimation, the
elevated ground objects such as buildings and trees or forest can
be 3D visualized from airborne SAR imagery. Herein, some
statistic algorithm such as Gibbs-Markov random fields can the
adopted for imagery segment. And, about the height estimation,
there will be three types of ground objects as follows:
- Elevated objects without crown, like buildings, woods
etc;
- Elevated objects with crown, like trees;
- The ground without clevated objects.
The height estimation methods of the first two cases have been
described in some rescarch (Kirscht, 1998), which must be
computed from the geometric relations between the flight
height, antenna location, objects location on the ground, the
corresponding SAR image coordinates, and the shape length of
the objects. In the last case, the height of ground without
elevated objects can be decide by the DEM in the spatial model
cell. Just like the above, after the two important factors, DEM
covered with orthoimages and buildings, have been extracted
from airborne SAR images, the vegetation distribution and
attributes information can be obtained from GIS basis graphics
database, even other factors such as traffic roads, buildings
foot-script and so on. Afterwards, the spatial reconstruction cell
can be available, and the flow chart is as follows (see as figure
1).
4.3 Model reconstruction
By 3D visualizing the buildings and vegetation and adding the
DEM covered with airborne SAR orthoimages, the
reconstructed spatial model with the time stamp of the special
day can be visualized by Virtual Reality technique. With the
multi-temporal stamp, it is difficult to visualize directly the
4.5D spatio-temporal model reconstruction cell, however, it can
be approximately described as a dynamic process like the
Figure 2.
Airborne :
SAR Imagery | GIS
images | segment data
Y 3 3
Ortho- Height | Buildings
images estimation foot-script
Y 1
DEM 3D visualizing
covered buildings and 1
with vegetation Vegetation
Ortho- distribution
images < |
Spatial Reconstruction model |
Figure I'S hy WR technique v chart from
T+0.5 A
^ Spatial
4 nodel
X Spatial
Z7 i »patia
ff model
; Spatial
: xdel
Spatial
‚model Y
Figure 2 the approximate 4.5D spatio-temporal model
Representation
74
In
Fr
an
fa
tre
fo
hij
hij
Tt
of
pr
re
int
pr:
ad
tin
lar
de
pp
su]
Int
Ap
pp.
El-
ten
Ad
Sle
