838
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
5. MATCHING ENTITIES
Until late eighties of the previous century, photogrammetric
measurements were still based on point-wise measuring
procedure. Point-wise approach requires that measured image
points in one image have to be identified on another image(s).
These point measurements must be conjugate and corresponds
to exactly the same 3D object point. The mathematical formula
for expressing image to object correspondence is based on
collinearity equation. The other approach of introducing higher
order primitives, usually referred as feature based assumes that
the measured points belong to certain object feature, which
fulfils a certain path in 3D space. The attractiveness of applying
feature based approach is that point to point correspondence is
no longer required. Control linear features can be used in all
photogrammetric procedures, in single image resection and
triangulation. The use of linear features is applicable in
mapping of either: man-made areas where plenty of buildings,
and road edges; or natural areas where river bank lines, coast
lines and vegetation boarders can be involved as linear features
(Heikkinen, 1992). Image space linear features are easier to
extract than point features. Meanwhile, object space linear
features can be directly derived form LMMS, GIS databases,
and/or existing maps, (Habib and Morgan, 2003).
Yet, the inclusion of linear features provide more constrained
solution for the exterior orientation parameters, and better
distortion modelling, however in some applications, is an option
like for example camera calibration. Meanwhile, the inclusion
of linear features becomes necessity, when they become the
only possible (realistic) similarity entities for example in case
of integrating LiDAR and photogrammetric data (Habib et al.,
2004b). Another example that demonstrates the importance of
involving linear feature constraints in photogrammetric
triangulation is the multi-sensor triangulation, where various
sensory data with different geometric resolution are involved.
In this case, the measurements of conjugate linear features
constitute a great advantage, for their ease identification in
different resolution images due to their high contrast (Habib et
al., 2004a). The linear feature constraint, see Figure 3, is
expressed mathematically as follows:
\V x ®V 2l
(2)
Similar to point features, linear features can be classified into
tie and control lines. Linear features, extracted from LiDAR
data, are control lines and can be used in different
photogrammetric operations. Linear features are useful to
relative orientation only when the same line is observed in an
image triplet (at least three images). In this case, the produced
planes will intersect in the object line indicating the quality of
fit.
The framework, introduced in this paper, involves linear feature
constrain since LMMS, with multi camera system, are usually
operated over road networks, where plenty of straight road
edges and lane line markings are available. The same features
will be mostly visible in AMMS images. Additionally, some
object space realistic constrains can be applied on linear
features like horizontal/vertical lines, and parallelism of object
lines. Downtown blocks have plenty of these constrains like
building edges and lane line markings. These constraints help in
the estimation of some of the georeferencing parameters. Thus,
linear features present major matching entities in the proposed
framework.
Figure 3: Linear Feature Constrain
The implementation of the linear feature constraint can be
implemented in a number of ways. The two vectors VI and V2
(in Figure 3) can be either replaced by their corresponding
object space vector that'joins the perspective centre and the
end/start point. Alternatively, they can be replaced by their
corresponding image space vector after back projecting the line
end points to the image space. Image space approach is much
efficient than the object space as it provides much higher
convergence rate than the object space cost function. This may
be attributed to the consistency of the system of equations with
those equations coming from collinearity constrain.
6. FRAMEWORK
The proposed framework for performing such integration
scenario is considered as a generic bundle adjustment that can
be easily extended to include additional matching entities and
also to add any functionality if necessary. In order to make the
developed framework fits the integration between mobile
mapping data, many features have to be added.
One of the basic modification, being a multi-camera enabled,
which is a must when involving land based data. This can be
done by involving another rotation matrix as a function of
boresight angles. Of course, any order of rotation is possible.
Regardless of the order, the three rotation angles are included as
parameters in the adjustment. A disadvantage of this procedure
is that the addition of these angles necessitates rather
fundamental changes to the implementation of the adjustment,
as the collinearity equations become functions of six angles
instead of just three. This, in turn, makes the linearization of
the collinearity equations considerably more complex.
However, the necessity of changing the adjustment model
presents a good opportunity to re-parameterise the
R M (rotation matrix between camera and mapping frame)
rotation matrix in terms of the roll, pitch, and azimuth angles.
This enables the values observed INS angles as well as their
covariance to be included in the adjustment.
Additionally, as compared to other optical-to-optical multi
resolution fusion, the proposed frame work involves linear
feature measurements implementation for reasons mentioned in
section 5. Another reason for using linear feature which is
tightly related to the land based mobile mapping system
bridging is that using linear feature has the advantage of
