ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002
from object space to sensor space in order to extract specific
features that might be needed during the inference process
to validate hypotheses, for example. Section 4 provides a
more detailed discussion.
2. BACKGROUND
In this section we briefly compare the advantages and dis-
advantages of the two most prominent methods for surface
reconstruction. We also elaborate on how to represent sur-
faces and emphasize the need for explicit surface descrip-
tions. Combining the advantages of LIDAR and stereo pho-
togrammetry is an interesting fusion problem.
2.1 Surface reconstruction
With surface reconstruction we refer to the process of deriv-
ing features in the 3D object space. The traditional method
of reconstructing surfaces is by photogrammetry. Here, a
feature on the ground, say a point or a linear feature, is re-
constructed from two overlapping aerial images—a process
known as stereopsis. This requires the identification of the
ground feature in both images as well as the exterior orien-
tation of the images. The crucial step in stereopsis is the
identification of the same ground feature, also referred to as
correspondence problem or image matching. Human oper-
ators are remarkably adept in finding conjugate (identical)
features. DEMs generated by operators on analytical plot-
ters or on softcopy workstations are of high quality but the
process is time and cost intensive. Thus, major research
efforts have been devoted to make stereopsis an automatic
process.
The success of automatic surface reconstruction from aerial
imagery is marginal. Despite considerable research ef-
forts there is no established and widely accepted method
that would generate surfaces in more complex settings, say
large-scale urban scenes, completely, accurately, and ro-
bustly. A human operator needs to be involved, at least on
the level of quality control and editing.
On the other hand, LIDAR has been touted as a promising
method to capture digital elevation data effectively and ac-
curately. LIDAR can be viewed as a system that samples
points of the reflective surface of the earth. The samples
are irregularly spaced. We call the original surface mea-
surements point cloud or mass points in this text. Laser
points are computed from navigation data (GPS/INS) and
range measurements. It is important to realize that there is
no inherent redundancy in the computation of a laser point.
In general, laser points do not carry semantic information
about the scene.
Table 1 lists a number of important advantages and draw-
backs of LIDAR and photogrammetry in terms of surface
reconstruction. LIDAR has a relatively high point density
and a high accuracy potential. However, in order to reach
the potential of a vertical accuracy of 1 dm and a horizontal
accuracy of a few decimeters, the LIDAR system must be
well calibrated. As is obvious from several accuracy stud-
ies, actual systems often do not yet reach this potential.
Recently, some LIDAR system offer the option to record the
entire waveform of the returning laser pulse. Waveform anal-
ysis yields additional information about the footprint area,
for example roughness and slope information. We have al-
ready elaborated on the disadvantages. Laser points are
positional, there is no additional scene information directly
Table 1: Advantages and disadvantages of lidar and aerial
imagery for surface reconstruction.
| | LIDAR aerial imagery 4]
ad- high point density rich in scene information
vant- high vertical accuracy | high H * V accuracy
ages waveform analysis redundant information
dis- no scene information stereo matching
ad- occluded areas occluded areas
vant- horizontal accuracy? degree of automation?
ages no inh. redundancy
available from a single point.
In contrast to laser points, surfaces derived from aerial im-
ages are potentially rich in scene information. Also, 3D fea-
tures in object space have a redundancy, r, of r = 2n 3
with n the number of images that show the same feature.
The Achilles heel of photogrammetry is the matching, that
is, finding corresponding features on nimages, wheren 2.
The degree of automation is directly related to the matching
problem.
From this brief comparison it is obvious that some of the
disadvantages of one method are offset by advantages of
the other method. This is precisely the major argument for
combining, or fusing if you wish, the two methods.
2.2 Implicit vs. explicit surface description
Common to the techniques of acquiring digital elevation data
is a cloud of 3D points on the visible surface. Except for
measuring DEMs on analytical plotters, the mass points are
irregularly distributed. Consequently, the next step is to in-
terpolate the raw points into a regular grid. Digital Elevation
Models (DEM) are immensely popular in many engineering
applications. With DEM we refer to a surface representation
with bare-earth z values at regularly spaced intervals in x
andy direction. A bare-earth DEM is void of vegetation and
man-made features—in contrast to a Digital Surface Model
(DSM) that depicts elevations of the top surfaces of features
elevated above the bare earth. Examples include buildings,
vegetation canopies, power lines, and towers. Finally, the
term Digital Terrain Model (DTM) is used as a DEM, aug-
mented with significant topographic features, such as break-
lines and characteristic points. A breakline is a linear feature
that describes a discontinuity in the surface. Such disconti-
nuities may signal abrupt changes in elevations across the
breakline, or may refer to abrupt changes of the surface nor-
mal.
Although very useful, it is important to realize that a DEM
or DSM does not make surface properties explicit. An ex-
plicit description of surface characteristics, such as planar
or higher-order surface patches, surface discontinuities and
surface roughness, is important for most subsequent tasks.
Object recognition and image understanding rely on the
knowledge of explicit surface properties and even the gen-
eration of orthophotos would greatly benefit from an explicit
description of the surface. In this line of thinking we con-
sider DEMs and DSMs as entirely implicit surface descrip-
tions and DTMs as partially explicit (if breaklines and distinct
elevation points are present). The challenge of describing
A- 311