‚ AB, T2N
algary.ca)
s remains to
orientation
he constant
ig to derive
ygrammetric
e extraction
features can
r, automatic
tinct points.
boundaries.
truction and
resented by
s automatic
stration, and
rections are
ortant factor
as surface
space, linear
maps, GIS
4MS).
with the use
:al functions
ygrammetric
)00). On the
| the use of
. The main
o establish a
's, whether
shapes, in
acquired by
ation should
's in multi-
nd other
single photo
scene/image
e motivation
| deals with
esentation ol
perspective
rimitives in
discussed in
ns for future
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
2. FEATURE TYPES
Features can be defined as discerned objects of interest that can
be extracted in the working space. Therefore, features in a
photogrammetric environment can be classified into image and
object space features. The following subsections discuss the
various types of features that can be found in the image and
object space.
2.1 Image Space Features
[mage space is two-dimensional. Therefore, primitives that can
be extracted from this space are either point, linear, or areal
features. It can be argued that areal features are the most general
features, because the image space can be classified into
homogeneous areal features/regions without leaving anv part of
g 2 g any p
the space unclassified. Linear and point features can be
considered as a subset of areal features whose dimension in one
or two directions is negligible because of scale or application
requirements.
2.2 Object Space Features
Object space is a three-dimensional space. Following the same
line of thought in the previous section, one can consider
volumetric features as the basis and most general primitives
capable of representing the entire object space. However,
appropriate object space features have to be based on the
photogrammetric procedure of object space reconstruction. In
photogrammetry, observing conjugate primitives in overlapping
images is utilized to derive corresponding object space
elements. The process usually starts by the extraction of image
space primitives, either manually or automatically. Then,
corresponding primitives need to be identified in overlapping
images. Finally, geometric attributes of the imaging system,
represented by the Interior and Exterior Orientation Parameters
(IOP & EOP), as well as matched primitives are used in an
intersection procedure to generate the corresponding object
space. Therefore, suitable object space primitives are those that
can be reliably extracted and matched in the input imagery. So
far, photogrammetric object space reconstruction has been
solely based on utilizing point and linear features. In other
words, there has not been an established procedure for object
space reconstruction using image space areal features. This
leads us to conclude that point and linear features are the most
suitable primitives for object space reconstruction from imagery
and consequently they should be the chosen primitives.
However, it should be noted that 3-D areal features (sometimes
called surface patches) could be considered when implementing
its dual representation, the bounding linear features.
23 Linear Features for Photogrammetric Activities: Why?
With the introduction and continuous evolution of digital
photogrammetry as a result of the availability of high-resolution
scanners and advances in digital image processing, automated
procedures are becoming popular. However, most of these
procedures are still point-based. Image processing techniques
allow for the extraction of numerous points with lower quality
than those selected manually. Moreover, automatic matching of
the extracted points in overlapping images is complicated and
often unreliable procedure due to variations in the imaging
System's point of view and relief displacements in the image
Space. Due to the above limitations, recent photogrammetric
research has been focusing on the use of linear features. This
line of research has been motivated by the following facts:
611
e Image space linear features are casier to extract when
compared to distinct points. This is attributed to the
nature of linear features since they represent
discontinuities in the grey value function in one
direction. On the other hand, point features represent
discontinuity in all directions.
* Linear features in the image space can be extracted
with sub-pixel accuracy across the direction of the
edge.
* [Images of man-made environments are rich with
linear features.
e Linear features allow for the incorporation of areal
features through the use of their boundaries.
Moreover, linear features are easier to use in change
detection applications than areal features. The
superiority of linear features stems from the
possibility of dividing them into smaller subsets. On
the other hand, breaking areal features into smaller
subsets is not a trivial task.
e Mobile Mapping Systems (MMS) can economically
provide accurate and current object space linear
features in real time.
e Linear features possess higher semantic information,
which are desirable for subsequent processes (such as
DEM generation, map compilation, and object
recognition). On the other hand, it is hard to derive
useful semantic information regarding the real world
from distinct points. Moreover, geometric constraints
are more likely to exist among linear features than
points. This will facilitate subsequent automatic
matching and object recognition activities.
e Linear features increase the redundancy and improve
the robustness and the geometric strength of various
photogrammetric adjustment activities.
Image space linear features can be either represented by an
analytical function (e.g., straight-lines and conic sections) or by
an irregular (free-form) shape. Among these representation
alternatives, one can argue that straight-line features are
appropriate for photogrammetric activities for the following
reasons:
e Man-made environments are rich with straight lines.
e Straight lines are easier to detect and the
correspondence problem between overlapping images
as well as between the image and object space
becomes easier.
a Straight-line parameters can be obtained with sub-
pixel accuracy.
e Free-form linear features can be represented with
sufficient accuracy as a sequence of straight-line
segments (poly-lines).
e Straight lines are valuable for the recovery of the IOP
of frame cameras, where object space straight lines
should appear as straight lines in the image space in
the absence of distortions. Therefore, deviation from
straightness in the image space can be attributed to
distortions (eg, radial and de-centring lens
distortions).
e Straight lines are valuable for the recovery of the EOP
of linear array scanners, where object space straight
lines will not appear as straight lines in the image
space due to perturbations (changes in the EOP) along
the flight trajectory.
