International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004
5. MATCHING
In order to automatically extract the DTM / DSMs from the
linear array images (airborne or spaceborne), algorithms and
software package developed in our group (see Gruen et al.,
2002 and Zhang et al., 2003) have been used.
The matching algorithm combines the matching results of the
feature points, grid points and edges. It uses a modified version
of the MPGC (Multi Photo Geometrically Constrained)
matching algorithm (see Gruen, 1985, Gruen et al., 1988 and
Baltsavias, 1991) and can achieve sub-pixel accuracy for all the
matched features.
Figure 7 shows the workflow of our image matching procedure.
| images and Orientation Data |
Image Pre-processing & Image
Pyramid Generation
; |
| Geometrically
Constrained
Feature Point | Grid Point Candidate Search,
Matching | | Edge Matching Matching Adaptive Matching
l Parameter
Determination
X
y
DSM (intermediate)
Combination of feature points, grid points and
edges
Modified Multi-image Geometrically
Constrained Matching (MPGC)
Final DSM
Figure 7. Workflow of the image matching procedure.
For the DSM/DEM generation, the SPOT-5/HRS images and
the previously triangulated orientation elements were used.
After the pre-processing of the imagery and production of the
image pyramid, the matches of three kinds of features, i.e.
feature points, grid points and edges, are found progressively in
all pyramid levels starting from the low-density features on the
images with the lowest resolution. A Triangular Irregular
Network (TIN) based DSM is constructed from the matched
features on each level of the pyramid and is used in turn in the
subsequent pyramid level as approximation for the adaptive
computation of the matching parameters. Finally the modified
MPGC matching is used to achieve more precise results for all
the matched features on the original resolution level (level 0)
and to identify some inaccurate and possible false matches. The
raster DTM / DSMs are interpolated from the original matching
results.
The main features of this matching procedure are:
e It is a combination of feature point, edge and grid point
matching. The grid point matching procedure uses
relaxation-based relational matching algorithm and can
bridge over the non- / little-texture areas through the local
smoothness constraints. Edges are introduced to control the
smoothness constraints in order to preserve the surface
discontinuities.
e The matching parameters include the size of the matching
window, the search distance and the threshold value for
cross-correlation and MPGC (Least Squares matching). For
instance, the procedure uses a smaller matching window, a
larger search distance and a smaller threshold value in
rougher terrain areas and vice versa. The roughness of the
terrain is computed from the approximate DSM on the
higher level of image pyramid. The adaptive determination
of the matching parameters results in higher success rate
and less false matches.
e Line features are important for preserving the surface
discontinuities. For this reason a robust edge matching
algorithm, which uses the adaptive matching window
determination through the analysis of the image contents
and local smoothness constraints along the edges, is
combined into our procedure.
eo Together with point features, edges (in 3D) are introduced
as breaklines when a TIN-based DSM is constructed in
order to provide good approximations for the matching on
the next pyramid level. The computation of the approximate
DSM for the highest-level image pyramid uses a matching
algorithm based on the iregion-growingi strategy (Otto et
al., 1988). According to this approach the already measured
GCPs and TPs are considered as i seed pointsi .
e The quality control procedure consists of (1) the local
analysis of the smoothness and consistence of the
intermediate DSM on each image pyramid level (2) the
analysis of the difference between the intermediate DSMs
and (3) the analysis of the MPGC results. Blunders can be
detected and deleted.
e For each matched feature, a reliability indicator is assigned.
Its value is based on the analysis of the matching statistics
(cross-correlation and MPGC results). As a consequence,
different weights are used during the generation of the grid-
based DSM/DEM.
Considering the characteristics of the SPOT-5/HRS image data,
some small modifications were introduced in our matching
procedure:
e The HRS imagery has 10 meters resolution in cross-track
direction and 5 meters in along-track direction (parallax
direction). This configuration may result in better accuracy
for point determination and DEM generation, but causes
some difficulties during the (area-based) matching
procedure. In order to avoid the problems, the images have
been resampled from 10m x 5m to 10m x 10m and
processed with our matching procedure (expect the MPGC
part). Then the MPGC (Least Squares matching) was run on
the original images in order to recover the original matching
accuracy. This two-step method results in the reduction of
the search distance between corresponding points, which is
equivalent to the reduction of the possibility of false
matching and the processing time.
e In some difficult areas, like small and steep geomorphologic
features (an example is shown in Figure 8), some manually
measured points can be introduced as iseed pointsi. This
operation gives better approximations for the matching.
iid
Figure 8. Manually measured seed points in difficult areas (two
small hills with steep slopes).
Inter
——
The
| incli
(rou;
| orde
softv
(feat
featu
mass
mate
Figur
