jes
)p-
rea
-T
3C-
4. DTM estimation
We use a module of the MATCH-T software package,
which is basically a robust estimation, to generate a
regular DTM grid from the filtered point cloud. This is
done for each of the overlapping DTM units separately.
The final step is the averaging of the overlapping areas,
which results in a DTM for the whole project area.
4.4 Results of first tests using the morphological filter
The above described automatic process for DTM generation
has been tested on a data set which was flown with 60/60
geometry, in terms of run time, disk space usage and
accuracy.
Table 1 summarizés the result of run times on each single
step of the DTM generation for a DTM shape, 92 by 92
mm in plate system. Note: Components marked with * have
do be done six times separately for each of the six different
combination of four images. The times shown in the table
are therefore six times a "normal" MATCH-T run.
was a conventionally measured surface description that rep-
resents the "real ground", namely a dense set of 3-D mass
points and break lines which were originally measured for ac-
curate contour interpolation. The table shows the RMS value
( Y — 25)2/(n — D) of the different data compared to
this reference. The difference in the RMS of the DTM com-
puted with and without the morphological filtering, confirms
the necessity of the filtering, especially in urban areas.
Accuracy
RMS
: [ft]
USGS DTM (level 1) 14.9
MATCH-T DTM without filtering (60um) 10.7
Filtered 3-D point cloud (60um) 3.4
Filtered 3-D point cloud (30um) 37
Final DTM using filtered 3-D point cloud (60pm) 3.4
Table 3:
Run time
DTM processing per 6 x 1/4 image
5 Final Conclusions of first results
At the time of writing the system had not been fully run from
beginning to end. However, the preliminary results of all the
different phases, particularly the AIO, the automatic mea-
surement of paneled control points, the filtering of the 3-D
point cloud, and the stereo editing system, have all been ex-
tremely gratifying. The system is expected to go into limited
production this summer.
REFERENCES
[FORSTNER W., GULCH E. 87] A Fast Operator for De-
Resolution 30pm 60pm
[min:sec] | [min:sec]
Normalization * 14:12 3:42
Image Pyramid * 5:42 1:30
Feature pyramid * 8:12 1:54
Point cloud generation (matching) * 15:24 5:48
Point cloud filtering 2:55 2:43
Final DTM estimation 6:17 6:03
| Total [..5242 | 210 |
Table 1:
Table 2 shows the disk space usage for MATCH-T run on
a single DTM shape model, and for the final (all four per-
spectives combined) DTM generation. Most of the data has
to be kept only temporarily. Only the final DTM and the
normalized images are needed for the final DTM editing.
Disk space usage
Resolution 30um | 60um
[MB] | [MB]
DTM processing per 1/4 image
DTM shape 18 5
Pyramid of the normalized DTM shape 23 6.6
Feature pyramid 10 2.6
DTM processing per 6 x 1/4 image
Point cloud (MATCH-T) 9 3.8
Final DTM (binary, 24 ft grid) 01 0.1
Table 2:
Table 3 summarizes the accuracy of different DTM's and
the 3-D point cloud. The reference for this investigation
tecting and Precise Location of Distinct Points, Corners
and Centers of Circular Features. Proceedings ISPRS In-
tercommission Workshop, Interlaken 1987, pp. 281-305
[FORSTNER W. 89] A Feature Based Correspondence Algo-
rithm for Image Matching and Least Squares Matching. In-
ternational Archives of Photogrammetry and Remote Sens-
ing, Vol. 26, Part 3, Rovaniemi, 1989
[FORSTNER W., WEIDNER U. 95] Towards Automatic
Building Extraction from High Resolution Digital Eleva-
tion Models. ISPRS Journal 50(4) pp. 38-49, 1995
[HARALICK R.M., STERNBERG S.R., ZHUANG, X. 87]
Image Analysis Using Mathematical Morphology. IEEE
T-PAMY, Vol 9, pp. 523 - 550, 1987.
[KRZYSTEK P. 91] Fully Automatic Measurement of Digital
Elevation Models. In: Proceedings of the 43rd Photogram-
metric Week, Stuttgart, pp 203 -214, 1991
[KRzvsTEK P., HEUCHEL T, HIRT U. 96] An Integral Ap-
proach To Automatic Aerial Triangulation and Automatic
DEM Generation. XVIII ISPRS, Commission Ill, Vienna,
1996
[POTH Z., SCHICKLER W. 96] The Automatic Interior Ori-
entation and its Daily use, XVIII ISPRS, Commission Ill,
Vienna, 1996
[WEIDNER U. 94] Parameterfree ^ Information-Preserving
Surface Rectoration. In: Eklundh J.-O. (Ed.), Computer
Vision -ECCV 94, Vol.11 Proceedings, pp. 218 - 224,
1994
873
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
