Hrabacek, Jan
configuration for the determination of an object point. Now, with sufficient constraints and well organised computing of
approximate coordinates, we can theoreticaly involve a point, which has no image line measurements related. However,
the extreme case when a point is determined by constraints only, we have not encountered yet. A more typical case
occured for Model #2, where only vertical lines were visible for lower features of the dorms. Approximate coordinates
were computed with the support of linear parallelogram constraints. A valid solution was reached by forcing the points to
the planes of the roof by means of the co-planarity constraints.
4.2.2 Enhanced computation of approximate values of coordinates Regression model (21) needs initial values for
all parameters in the model. Our approach uses exterior orientation parameters approximately known, determined in a
separate step. Then the object point coordinates and the object plane parameters remain to be determined. For the points
related to a sufficient number of diversely oriented line observations, approximate values of object points are computed
by the intersection of interpretation planes. In following steps of the iterative process, points are inresected also using
preliminary determined object planes.
Besides the announced possibility of modelling with the use of a priory geometric information, this information can also
be used for computation of initial values. in the presented models, the initial values of exterior orientation were separately
computed outside the adjustment software. Computation of approximate coordinates of weak points was supported with
the use of parallelogram constraints, where this constraint was relevant. Although not yet implemented, other constraints
can also support the computation of initial values of object points for which less than three interpretation planes are
available.
5 CONCLUSION
The integration of weighted constraints in a line-photogrammetric bundle adjustment is presented. A new solution to
the weighted parallelism constraint on planes has been developed. Based on the presented approach, we developed a
tool integrating adjustment and constraint-based modelling. With this approach we are able to reconstruct models using
a geometrically weak set of image line observations, with exterior orientation parameters only approximately known.
Despite the weakness and occlussions, a regular shape of the model can be obtained. Furthermore, the number of images
needed is minimised.
Regarding our experiments, the used weights provide the convergence and the desired regularity. However, low values of
residuals of constraint observations (so-called pseudo-observations) indicate that an optimal balance of weights of image
observations and the pseudo-observations is still unclear and needs further investigation. Automation of some steps of
processing is one of the possible future improvements. Semi-automatic creation of topology, based on automatically
extracted image lines is an example of such an improvement. An efficient tool was developed for the reconstruction of
valid, regularly shaped models suitable for an architectural visualisation. An integration of all steps of processing into a
common environment would make the tool even more efficient.
REFERENCES
CIPA, 1999. Zürich City Hall. http://cipa.uibk.ac.at/dataset.html. A reference data set for digital close-range photogram-
metry.
Mikhail, E. M., 1976. Observations and Least Squares. EP-A Dun-Donnelley Publisher, New York.
Rektorys, K. (ed.), 1994. Survey of Applicable Mathematics. Vol. 1, 2"? revised edn, Kluwer Academic Publisher,
Dordrecht.
van den Heuvel, FE. A., 1997. Exterior orientation using coplanar parallel lines. In: 10** Scandinavian Conference on
Image Analysis, Lappeenranta (Finland), pp. 71-78.
van den Heuvel, F. A., 1998. Vanishing point detection for architectural photogrammetry. In: International Archives of
Photogrammetry and Remote Sensing, Vol. 32, Hakodate, pp. 652-659. part 5.
van den Heuvel, F. A., 1999. A line-photogrammetric mathematical model for the reconstruction of polyhedral objects.
In: F. S. El-Hakim (ed.), Videometrics VI, Vol. 3641, SPIE, pp. 60-71.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 387