Mohammad Saadat Seresht
Other techniques of 3D-line extraction are line matching. In this case, we can derive approximate parallax
of stereo lines from DSM.
5 BUILDING RECONSTRUCTION
After coarse recognition of building regions and fine recognition of building details, reconstruction step is
started. Before reconstruction step, extracted features have not been depended together and can not be
supposed an object named, building. But after reconstruction step, these features as well as DSM data and
other exterior information (e.g. knowledge about shape of building or using of side images) can be built a
representation of building.
Definition of reconstruction is a 3D geometric description of building in requirement level of details. This
3D geometric description needs 3D geometric modeling that has several methods (figure 1).
Geometric Modelling
| Graphic-Based Modelling | | Surface-Based Modelling | | Solid-Based Modelling |
|
[Non-Composition Mode! | | Constructive Model | Boundary Model
Figure 1. Types of geometric modeling
In building reconstruction we have several methods that are related to following parameters:
Building types: From simple rectangle with flat floor to much more complicated building shape
Level of details: From building blocks to single buildings with ridge, gable, chimney and dormer
Number of Images: From stereo images to multiple images
Type of image cue: Such as color, texture, or shadow and reflectance
Primitives: Such as points, 2-leg or 3-leg vertex, lines and homogeneous regions
Existence information: Such as other sensors data, DTMs, DSMs, maps or GIS data
Automation level: From manual to full automatic
Processing space: From full 3D processing to composition of 2D and 3D processing
Geometric conditions: Such as pre-knowledge about building, its geometry and functionality e.g. right
angle, parallel/straight lines, sensor geometry, sun location and so on
Totally in reconstruction step, geometric models either are predefined e.g. generic models or are created
directly by extracted features in required level of details. No specific model of geometric modeling does it
In relation with this, DSM id very good help for approximate modeling that can be used for starting point in
reconstruction and therefore it improves reliability of reconstruction.
At last it note be that, because building have different shape and complexity, operators don’t have skill on
understanding of system and density and complexity of building regions, building can not be reconstructed
completely. Therefore it is better that, in addition to using of above algorithms, a method of control is
applied e.g. simultaneously 3D viewing of scene as well as reconstructed building representation.
6 EXPERIMENTS
Input data are large-scale stereo images (1:3000 scale) and DSM with height intervals of 1 and 5 meters
respectively and output is linear features of each building to be used in reconstruction step.
In figure 2, output of each section is represented: a) sample of building region b) related DSM c) automatic
classification of DSM by seed filling method d) using of building recognition image filters and fining 3D
region boundaries e) projecting up 3D region boundaries on images by collinearity condition f) edge
796 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
extr:
exist
that «
and s
probl
occlu
solve
are nc
fusioi
more
enabl
|
availa
r
achie
ruin t
may b
REFI
Balts:
Extra
—
