The mathematical model of the bipartite weighted matching
algorithm can be expressed as: We find out the optimal
matching
+ + +
* * l l res NE
$1] -( i 7 E (2.10)
JI sai wot
such that
Mi bs :(LJ) eb, - T 2.11
ax ce )e Fm (2.11)
Now this model is used by pairs of scan lines. The application
in the whole image at one time should be possible.
3. Experimental results
3.1 Results of automatic point transfer
The old image set was taken in 1989 with Wild camera RC10.
The focal length is 213.91 mm. Photoscal is 1:3700 and
scanned by Vexel scanner with resolution of 20 um. The new
image set was photographied in 1992 by Zeiss RMK-TOP
30/23 camera at about 1000 m high. The focal length is 305.412
mm. The images were also digitized by Vexel scanner with
1200 dpi.
After necessary preparation (automatic interior orientation and
image resampling to epipolar geometry), the operator should
choose the positions which has more features in the image.
Anyway if there are great differences between the two image
sets, the results will be very poor even no back position.
Fig.3.1.1 gives out one example
Fig.3.1.1 The two image sets with found point '*'
3.2 Results of DTM production
In order to check the robust ability of the of the algorithm,
some tattered areas were chosen. Fig.3.2.1 and Fig.3.2.2 show
some area in quarry field and its matching results. The matched
points are marked with white points.
Unlike automatic point transfer, the returned matching point
will be displayed. The operator can check them one by one with
stereo viewing. By DTM computation, in order to have a
qualitative checking, the digital terrain model of the same area
was measured manually. Or like that in our case some amount
of well-defined positions are chosen and checked on analytical
plotter. In this example thirty well-defined feature points were
picked out from both models for the comparison. The table
3.2.1 shows the checking results.
986
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
Table 3.2.1
Fig.3.2.2 Matching results in mixed area
4. Discussion
4.] About automatic point transfer
At presents there are some limitations about this procedure.
First of all the positions of project centers of the same image
areas between two flights should not be very far away. We have
that for our example about 200 ~ 300 m. For the new flights it
should not be far away of + 80 m from each other. Beside this
the flight height, the flight date and time, the camera of the two
aerial photographies should be similar as much as possible. The
resolutions of digital images can be the same, can also be
differently.
Such ‚a technique has a very bright future for daily
photogrammetric production. As in our case, it can save
aerotriangulation in every 3 ~ 4 years without any accuracy
loss.
4.2 About DTM production
The matching method described in this paper is suitable for the
different types of aerial photos, different image contents except
that image with no features. Due to the fact that only those local
geometrical and radiometric noise independent properties have
been used as matching representations, even though the stereo
pair are quite different, the global bipartite weighted matching
can also get satisfied results. Here we have to point out that the
noise means not only the real one but also those image