Line-photogrammetric approach with weighted constraints in an architectural application
Czech version of the document
About the project:
The contribution to a research of line-oriented photogrammetrical processing has been realized by Jan Hrabacek under supervision of Frank van den Heuvel within a three month internship (9/99-12/99). The project was focused on extending line-photogrammetric model and is described in [Hra2000] . The internship has resulted in an extension of the software and processing of a test data set using the software.
Objectives: line-photogrammetric principle (&& software) extensions
Geometrical constraints are used to improve the quality of the reconstruction that has been obtained using line-photogrammetric principles only. Geometric constraints represent information about geometrical relations between object features. Examples of these constraints are listed below.
The following types of geometrical constraints are implemented in the research software (the last two by Jan Hrabacek):
- Angle constraint - assumes a known angle between two planes of an object. This constraint is mostly used to enforce perpendicularity between two planes.
- Parallelism constraint - the previous constraint cannot be used to set two planes parallel. Therefore, a new original solution was developed, employing a pair of constraints.
- Plane-plane-distance constraint - assumes a known distance between two parallel planes to be known. It can be used when the size of a building has been measured directly.
Weighting of constraints
is an extension that made the approach more powerful. Weighting allows to account for uncertainty in the geometric constraints due to inaccuracies in construction, or due to deformations of a building.
Test data set:
As
a test data set,
the CIPA data set
of
Zürich City Hall has been used. This data set is intended to become a reference data set for comparing and testing of digital photogrammetry approaches and methods. The set contains two subsets of images taken by two different cameras, and a detailed description including control points information. For more details, consult
Facts:
- only 4 images used (of 15 possible)
- camera: Olympus C1400L
- size of images: 1280 x 1024 pixels.
Models:
Models have been computed using the extended line-photogrammetric software.
The first and the second model have been reconstructed without small roof parts (dorms) and a comparison between them demonstrates the potential of the constraint-based reconstruction.
The third model involves all dorms, which are mostly poorly visible and are partly occluded in the images.
For all these adjusted models only approximate values of exterior orientation parameters were used, and no externally measured control points.
Models have been converted to VRMLs, some of them contain a visualization of error ellipsoids. Texture mapping tries to make the models more realistic. However, nice texture mapping and the use of a small number of images, are in a certain contradiction. Here, the minimization of the number of images is regarded as an important goal of the research.
For each model, images and VRML models are presented below. To view the VRML models, the plug-in CosmoPlayer can be used.
(without any constraints)
- # of observation: 112 lines
- # of unknowns: 154 (18x points x3 + 18x planes x4 + 4x exterior orientation parameters x7)
- # of geometrical constraints: 0
<table class="kader" width="75%"><tbody><tr valign="top">
image of the model
(5 kB)VRML model <td/></tr></tbody></table>Both the walls and the roof do not have a regular shape (Computation done without constraint support)
(2.5 kB)
(with constraints)
- # of observation: 112 lines
- # of unknowns: 154 (18x points x3 + 18x planes x4 + 4x exterior orientation parameters x7)
- # of constraints: 22
image of the model
(5 kB)VRML model
(2.5 kB)VRML model with
texture mapping
(130 kB)
Significant improvement of the shape regularity (Walls perpendicular, roof of a regular shape)
Model #3 (the most complete model with constraints)
- # of observation: 393 lines
- # of unknowns: 434 (74x points x3 + 46x planes x4+ 4x exterior orientation parameters x7)
- # of constraints: 52
image of the model
(9 kB)VRML model
(10 kB)VRML model with
visualisation of error ellipsoids
(26 kB)VRML model with
texture mapping
(130 kB)
By further constraining the model, the dorms can be reconstructed (parallelism & perpendicularity & coplanarity relations)
References, addresses & URLs:
Czech Technical University in Prague
Department of Advanced Geodesy e-mail:
Delft University of Technology
Department of Geodetic Engineering
Section of Photogrammetry and Remote Sensing email:
F.A.vandenHeuvel@geo.tudelft.nl
[Hra2000]
Hrabacek, J. and F.A. van den Heuvel, 2000, Weighted Geometric Object Constrainst
Intergrated in a Line-photogrammetric Bundle Adjustment
In: Proceedings of the XIXth congress of ISPRS, Amsterdam 2000,
"International Archives of Photogrammetry and Remote Sensing", vol XXXIII, part B 5/1, commission V, ISSN 0256-1840, pp. 380-387.
[download the pdf-file]
Last update: January, 2000
Frank van den Heuvel
