SINGLE IMAGE AND TOPOLOGY APPROACHES FOR MODELING BUILDINGS
O. Al Khalil & P. Grussenmeyer
ENSAIS - Photogrammetry and Geomatics Group, 24 Boulevard de la Victoire, 67084 STRASBOURG Cedex, France
alkhalil@ensais2.u-strasbg.fr pierre.grussenmeyer@ensais.u-strasbg.fr
Commission V, WG V/2
KEY WORDS: Geometry, Photogrammetry, Reconstruction, Modeling, Building, Database, Image
ABSTRACT:
The paper presents a method and strategy for building reconstruction and visualization from single images and topologic approaches.
In our method, geometric and semantic data of entities to be modeled are extracted. The structuring of geometric, topologic and
semantic data is done within the measuring process.
In most applications, the model contains only information about the shape and the texture of the object while its topologic and
semantic data are not processed. We combine this information to reconstruct outdoor and indoor scenes.
Single image techniques based on projective geometry aspects as vanishing points and homography are adopted. This configuration
was preferred to the traditional stereoscopic and multi-image techniques considering the particularity of the environment to be
modeled and application difficulties of these techniques.
Indoor scenes are structured in blocks like rooms, corridors, etc. with their own orthogonal coordinate systems. A block is presented
in an abstract way using a Data Conceptual Model to identify the entities and the relations to include in the database. The 3D model
can then be reconstructed automatically from the information stored in the database. Coplanarity conditions between entities as
windows or doors and the compatible walls are taken into account. The outdoor and indoor blocks are merged by the use of “virtual
doors” (or windows). These doors are defined in function of the wall thickness and the translation direction according to the local
coordinate systems of the blocks. The final model offers several applications:
- topologic navigation in the model,
- visualization and extraction of the geometric characteristics of given types of entity (areas, perimeters, etc.),
- visualization of " virtual doors ", used to calculate the parameters of the transformation (rotation and translation) between the
reference coordinate system and the local systems attached to the different blocks,
- generation of photo-realistic models
As examples, the method has been applied to the modeling of the exterior part of the old Zurich city hall (CIPA dataset from 1999)
and to the modeling of a part of the ENSAIS buildings.
1. INTRODUCTION
Building modeling methods are of interest for urban planning
and for projects in civil engineering. These methods consist in
reconstructing the exterior parts of buildings: roofs and façades.
The 3D geometric data are usually extracted using stereoscopic
or convergent images rays.
The goal of this paper is to present a strategy for 3D indoor
modeling based on the combination of single image
photogrammetry and topologic approaches. In this method, data
structuring is done within the measurement process.
Using single image configuration allows to:
- reduce the number of images;
- avoid the establishment of conditions required for the
traditional approaches (stereoscopic and convergent
images rays);
- use archive images.
Our test object for indoor modeling is a building at ENSAIS.
Each floor of the building is structured in a reference
component (the corridor) to which simple components are
attached (classroom and offices) by interconnection surfaces
(doors). These surfaces were used to merge the different
components.
Although the proposed strategy is applicable to indoor scenes, it
can also be applied to outdoor ones. To show this possibility,
our method was used to establish the 3D model of the old
Zurich City Hall building from a dataset proposed by CIPA.
2. DESCRIPTION OF THE METHOD
Our research is in the field of 3D modeling and representation
of indoor scenes. This approach is based on:
1. The organization of the indoor part as being composed of
simple unit components or reference unit components;
2. The representation of the information in hierarchical form;
3. The simultaneous structuring of data in a relational
database.
The 3D geometric data are obtained from single images. Data
structuring, which includes the geometry, the topology and the
semantic aspects of the modeled scenes, is done during the
geometric modeling process.
The Conceptual Data Model (CDM) represents the information
in three geometric, semantic and topologic levels.
The geometric level of modeling (and its relations with the
topology), is a graphic representation, which only reports the
position and the shape of objects and the spatial relations
between these objects (neighborhood, intersection, slope, etc.).
The topologic level is an intermediate level used to define the
relations between geometric primitives (points, lines and
surfaces) presented in the previous level. The topology allows
an effective structuring of the graphic elements. In this context,
a point represents a nod, a segment represents an edge, a surface
represents a face and the unit component itself represents a 3D
body.
—131—
