parameters of the panoramas (X,Y,Z,) differ significantly
from the approximate values. Standard bundle adjustment
would often fail for such configurations. Since tie points can be
distributed all over the horizon of a panorama, the here
presented approach is reliable and robust. Up to now there is no
experimental set-up where the process failed. A remarkable
advantage of this approach is the limited number of required tie
points. Recent investigations have shown that a total of 5 to 7
points is sufficient for a complete room.
4.3 Achieved accuracies
The achieved accuracies asre mainly based on the following
criteria:
= quality of camera calibration
= quality of tie point measurement for panorama generation
= quality of tie point measurement for panorama orientation
= distribution of panoramas inside the room
In order to assess the first three topics theodolite measurements
of control points (better than 1mm) have been carried out.
Subsequently the whole process has been tested based on
ellipse-operator measurements of circular targets on one hand,
and manual cursor measurements of natural points on the other
hand. The observed object space has dimensions of 14m x 12m
x 2.5m. Tab. 1 summarizes the deviations of adjusted object
coordinates with respect to the theodolite results.
circular targets natural points
mean deviation 21/25/03 10,1/98/2,53
max. positive deviation 31/68/19 16,5 / 16,7 / 4,2
max. negative deviation -7,3/-7,4/-1,6 | -15,3 / -9,6 / -2.0
Tab. 1: Deviations of adjusted coordinates compared to
theodolite measurements (X/Y/Z, in mm)
Compared to signalised targets the use of natural tie points
leads to worse results, mainly due to the limited detectability of
those image points. However, for applications with less
accuracy specifications even natural tie points may yield
sufficient results. Since no targeting is used the effort for object
preparation and in-house measurements is drastically reduced.
Besides image acquisition only a few (minimum 1) object
distances have to be measured on-site.
5. 3-D OBJECT RECONSTRUCTION
After orientation object points can be measured by spatial
intersection. Usually edge and corner points of object features
are used as object points. Since there is no fully-integrated
program environment available, measured object points are
transfered into a 3-D CAD system such as MicroStation where
the final modeling is performed. Fig. 15 shows an example of a
3-D room model with interior furniture.
6. SUMMARY AND OUTLOOK
The presented approach for generation and use of panoramas is
an efficient and cost-effective method for the reconstruction of
3-D object information with photogrammetric precision. While,
for interior room surveying, the standard multi-image approach
of close-range photogrammetry leads to a relative high number
of images with rather weak configuration, only 3 or 4
panoramas are required. If only room inspection is desired, only
one panorama would be sufficient. A small number of tie points
has to measured, usually only 5 to 7 points.
Future works will concentrate on the implementation of
automatic matching algorithms for the stitching process of
adjacent images. This step should increase accuracy by using a
higher number of tie points whereby manual interaction are
reduced.
In addition, the mathematical orientation model will be refined
in order to provide higher accuracies for object reconstruction.
An integrated program environment is currently under
development.
Fig. 15: Example 3-D model made by panorama object
reconstruction; left: measured object information; right:
rendered 3-D view
7. REFERENCES
Düppe, R.-D. (1998) Beispiele zur Umbildung von
Weitwinkel-, Panorama- und Fisheyeaufnahmen im
Nahbereich. Allgemeine Vermessungsnachrichten, Karlsruhe.
Hóhle, J. (1998): On the production of photorealistic and
dynamic 3D-models of building structures by means of digital
photogrammetry, Proceedings Visual Reality, pp. 141-150.
Hóhle, J. & Pomaska, G. (1999): Zur Visualisierung von
Gebàudemodellen und deren dynamische Repräsentation im
Internet, http:/www.imagefact.de/sokrates/d/jhgp.html
Lisowski, W. & Wiedemann, A. (1999): Auswertung von
Bilddaten eines Rotationszeilen-scanners. Publikationen der
DGPF, Band 7, pp. 183-189.
Pomaska, G. (1998): Automated processing of digital image
datat in architectural surveying. IAPRS, Commission V,
Hakodate, Japan or http:/www.imagefact.com
Scheele, M, Borner, A., Reulke, R., Scheibe, K. (2001):
Geometrische Korrekturen: Vom Flugzeugscanner zur
Nahbereichskamera. Photogrammetrie-Fernerkundung-Geoin-
ormation, Heft 1, 2001, pp. 13-22.
—186—
ABS
AsT
mor
appl
for
We
crea
poin
witk
the |
such
Twi
cult
Bot
AS |
bec
bec
can
geo
asc
hav
dig
app
usii
usu
exp
cor
all
gec
pro
im:
wit
on
nol
en
im:
pei
