The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008
we introduced EDISON detection algorithm to color remote
sensing images. As reported recently, EDISON works well in
edge detection of images used in other fields. This algorithm
proposes an image segmentation and confidence based edge
detection model.
Besides the above, this edge model has other advantages. The
first is a lack of false negatives compared to other models false
negatives result from a failure to “take into account all possible
intensity variations that might accompany a step edge in
practice”. Almost all of these variations are modelled implicitly.
In homogeneities can be uncorrelated (due to noise) or
correlated (due to texture) without affecting performance. The
second benefit is that using distributions creates a unifying
framework for edge detection in binary, grey-scale, color, or
multi-spectral images, so long as a meaningful ground distance
is defined.
2.2 Dense Stereo Match
In order to generate DSM, under a energy minimization
framework, a dense stereo match between stereopair is taken. In
the case of dense stereo match. The concept of a disparity space
image or DSI is important. In general, a DSI is any image or
function defined over a continuous or discretized version of
disparity space (x, y, d). (Scharstein, D. and Szeliski, 2002)In
practice, the DSI usually represents the confidence or log
likelihood (i.e., cost) of a particular match implied by d(x, y).
The goal of a stereo correspondence algorithm is then to
produce a univalued function in disparity space d(x, y) that best
describes the shape of the surfaces in the scene. This can be
viewed as finding a surface embedded in the disparity space
image that has some optimality property, such as lowest cost
and best (piecewise) smoothness. So a stereo correspondence
can be formed to a energey minimization framework.
Under the DSI, we test various energy minimization algorithm
and find among the common ones, graph cut works best but
causes most computation cost. In order to get the best DSM
data, we use In this case, we preprocess the images to generate
DSM and save to DSM database. Therefore, the huge
computation cost has less effect on the efficiency of our system.
3. BUILDING MODEL DRIVING (BMD) ALGORITHM
3.1 Geometric building model
For the task of building reconstruction, we first of all have to
define the geometric building model. This model defines the
hip-roof building and box building, which can be dealt with.
(Brenner) comes to the conclusion, that the combined
parametric models were the most suitable ones for automatic or
semiautomatic building reconstruction. Both provide a good
compromise between the reconstruction effort and the number
of supplied real world building types. Combined parametric
models imply geometric and topologic characteristics. They are
based on constructive solid geometry (CSG). A large number of
different building types can be modeled, especially buildings
with different eave heights. In this paper, due to the low
quantity of the DSM, we only describe the hip-roof building
and box building model.
Figure 1: Geometric modal of the two kinds of building in this
paper
Object recognition or reconstruction, in general, presumes
knowledge about the perceived objects by some kind of object
model. These object models can be considered as abstractions
of real world objects. As model definition it is important to find
balance between correctness and tractability i.e. the results
given by the model must be adequate both in terms of the
solution attained and the cost to attain the solution (Streilein,
1996). A priori knowledge or in other words constraints can be
introduced by applying a very rigid building model. A rigid
building model restricts the search space, which has to be
examined to find a solution. On the other hand these models
limit the number of possible building types which can be
represented by a single model. In order to deal with the large
architectural variations of building shapes, the utilized model
should be as general as possible. Since most buildings are
bounded by a set of planar surfaces and straight lines, in our
approach a building is represented by a general polyhedron.
Additional constraints are defined by the assumption that the
coordinates of the given ground plan are correct and the borders
of the roof are exactly defined by this ground plan. This
provides sufficient restrictions to enable the reconstruction of
buildings without loosing the possibility to deal with very
complex buildings.
Two types of representation are feasible to describe the
reconstructed buildings. The boundary representation (BRep) is
probably the most widespread type of 3D representation. Many
algorithms are available for computing physical properties or
visualizations from that representation. The object is
represented by its surface, which is decomposed into a set of
faces, edges and vertices. The topology is additionally
described by a set of relations which indicate how the faces,
edges and vertices are connected to each other. In constructive
solid geometry (CSG) simple primitives are combined by
means of Boolean set operators. A CSG representation always
results in valid 3D objects, i.e. in contrast to a BRep no
topological check has to be performed in order to guarantee the
closeness of the object surface.
3.2 Model Selection and Parameter Estimation
Each building is described by a combination of one or more
basic building primitives. Each of them consists of a cuboid
element with different roof types flat roof and hip roof. This
type of representation is similar to the one used by (Englert and
G 'ulch, 1996). In order to reconstruct more complex buildings,
first the complete building has to be decomposed into these
basic structures. This step can be realized automatically by the
analysis of the given ground plan. Figure 6 shows the result of a
ground plan decomposition into rectangular structures. Every
rectangle defines one building primitive. Since position,
406
