ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002
collinearity equations and uses different forms of trajectory
models. This sensor model is used for the improvement of the
measured exterior orientation parameters for each scan line of
TLS images by a modified photogrammetric bundle adjustment,
and for the derivation of the geometric constraints in our
modified MPGC and GCMM procedures. More details on our
TLS sensor model can be found in Gruen and Zhang, 2002.
Table 1: TLS sensor and imaging parameters
focal length 60.0 mm
number of pixels per array 10200
pixel size 7 um
number of CCD focal plane arrays 3
stereo view angle 21/42 degree*
Field of view 61.5 degree
instantaneous field of view 0.0065 degree
scan line frequency 500 HZ
* forward-nadir/forward-backward stereo view angle
View from the Lens
CCD Flight Direction
Scanning EPI »
Direction
| OCDE GEDO CODI.
«o =»
B G R B G R Same Structure
10200 10200
A A
| | 71.4mm
| 7 7 7 1 7 7 M
Ip po rép UA U
(lere ie — Ferr
n I n 1
' 1544 1540 | 154. 154m '
23.032mm Ld
Figure 1: TLS CCD sensor configuration
3. Matching Considerations
The automatic generation of DTMs has gained much attention
in the past years. A wide variety of approaches have been
developed, and automatic DTM generation packages are in the
meanwhile commercially available on several digital
photogrammetric workstations. Although the algorithms and the
matching strategies used may differ from each other, the
accuracy performance and the problems encountered are very
similar in the major systems and the performance of commercial
image matchers does by far not live up to the standards set by
manual measurements (Gruen et al., 2000). The main problems
in DTM generation are encountered with
(a) Little or no texture
(b) Distinct object discontinuities
(c) Local object patch is no planar face in sufficient
approximation
(d) Repetitive objects
(e) Occlusions
(f) Moving objects, incl. shadows
(g) Multi-layered and transparent objects
(h) Radiometric artifacts, like specular reflections and others
(1) Reduction from DSM to DTM
The degree to which these problems will influence the matching
results is image scale dependent. A DTM derived from 10m
pixelsize SPOT images will be relatively better than one derived
from 10cm pixelsize TLS images.
Area-based, feature-based and relational matching have both
advantages and disadvantages with respect to these problems.
The key to successful matching is an appropriate matching
strategy, making use of all available and explicit knowledge
concerning sensor model, network structure and image content.
But even then the lack in image understanding capability will
lead to problems, whose relevance must be judged by the
project specifications.
This paper presents a matching procedure for automatic DSM
generation from the TLS raw images that can provide dense,
precise and reliable results and addresses the problems (a)-(f)
mentioned above. The proposed method is a combined
matching procedure, which is based on both grid point matching
and feature point matching. The presented results reflect an
intermediate stage of development. We are fully aware that
more refinements are needed before automated matching can be
considered a highly reliable procedure.
Figure 2 shows the strategy of our matching approach. We use
the raw TLS images and the given or previously triangulated
orientation elements. After production of the image pyramids
we extract on the upper pyramid level a first approximation
DSM by a geometrically constrained feature point matching
based on cross-correlation. Next we run a grid point based
relaxation matching scheme through all pyramid levels. We
select the grid width to 11 pixels on all levels. The matching
candidates are obtained by a cross-correlation-based
geometrically constrained matching, as described in chapter 4.1.
At each level we obtain a refined DSM, which in turn is used in
the subsequent pyramid level for the candidate search. The
important aspect of this relaxation method is its compatible
coefficient function and its smoothness constraint satisfaction
scheme. The smoothness constraint links the matching results of
the neighbouring grid points to each other and achieves global
consistency in the matching. The weight of the smoothness
constraint is related to the image texture information and
provides the possibility of controlling the continuity of the
terrain surface. With the smoothness constraint, image areas
with little or no texture information can be bridged by assuming
that the terrain surface varies smoothly over the area.
Next we can either activate a modified Multiphoto
Geometrically Constrained Matcher (MPGC) or a
Geometrically Constrained Multi-point Matcher (GCMM). Both
may be considered as refinements of the relaxation matching
results and are used in order to achieve sub-pixel accuracy.
The modified multi-point matching with geometric constraints
is characterized by its smoothness constraints in the 2D parallax
domain. By including geometrical constraints, it can be used to
match three TLS images simultaneously and provide the pixel
and object coordinates for each nadir image grid point
simultaneously.
In order to compensate the disadvantages of terrain modeling by
grid points, a feature point matching procedure which exploits
the modified MPGC has also been implemented. In this
procedure, the feature points are extracted by using some
interest operator such as Moravec’s. The relaxation matching
results provide quite good approximations.
The weighted geometric constraints in the modified MPGC and
GCMM forces the matching to search for a conjugate point only
along the epipolar curves. This reduction of the search space
from 2D to 1D increases the success rate and reliability of the
feature point matching results. Moreover, the geometric
constraints derived from the TLS sensor model link the grid
matching results of the three TLS images and have the ability to
solve the problem of repetitive objects, occlusions and moving
objects in image matching.
According to Hsia and Newton, 1999, using a combination of
feature points, grid points and filling back points can give
encouraging results for DTM production. The results of using
three different TLS data sets from Japan (city, sub-urban and
mountainous area) will be reported in this paper.
A- 132