9-17 Nov. 1999
*
..
°
e
e
au^ .
.* ns
°
.
°
se.
°
e
0. $ t
Sees
Shel
e
Po
© $99
1
°
°
e principle of generating
ta set. Based on an as-
tic laser images are com-
ically on stereoscopes, an-
ations. The latter possibil-
se through large data sets
(b), a stereogram of laser
apartment building shown
pe, a vivid 3-D impression
appears.
fer to the raw laser sur-
erties, such as smooth
object boundaries. The
erties is expressed by a
nds on the nature of the
the semantic aspects of
isidered. The following
tric errors of points and
nts A simple but most
metric accuracy of raw
m with known surfaces.
except perhaps for cal-
her ways to assess the
ser footprints offer the
ysis. This internal anal-
physical surfaces are in
sis is usually performed
ee, e.g. Csathô et al.
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999
(1995)). Clearly, the errors derived in this manner are
relative, reflecting the inner accuracy of the laser rang-
ing system.
A better way to assess the geometric quality of a laser
data set is to compare it with a description of the same
surface but obtained from an independent source, for
example with photogrammetry. Laser ranging generates
an irregularly distributed set of surface points. It is very
unlikely that the control surface is represented by the
same points. Thus, the challenge is to compare two dis-
crete surface descriptions that have no identical points.
Converting both data sets to a regular grid is a popu-
lar method but the comparison of grid post elevations is
now affected by interpolation errors. Another restriction
is that both sets need to be in the same reference frame.
A better solution is to compare the differences between
the two sets at their original point locations. This is pos-
sible except that a local surface patch must be generated
for one of the two sets. Details of such an approach are
discussed in Habib and Schenk (1999) and Postolov et
al. (1999), for example.
Quality Control of Derived Surface Properties The
error analysis of raw laser points yields important in-
formation but does not tell much about the accuracy
of derived surface properties, such as breaklines and
smooth surface patches. The error analysis of derived
surface properties does not only depend on the inher-
ent quality of surface points, obtained from laser rang-
ing or photogrammetry, but also on the feature extrac-
tion methods. In fact, post-processing algorithms of-
ten have a larger impact on the quality of extracted fea-
tures than the raw surface points. If the extracted fea-
tures have known geometric aspects then a simple qual-
ity control measure is to determine the difference be-
tween extracted and known properties. Breaklines may
be straight lines, for example; in addition they may be
horizontal and have known lengths. Smooth surface
patches may be planar, and have perhaps even known tilt
angles. If no control information exists about extracted
features then one can measure it photogrammetrically
in order to check the feasibility of the extraction algo-
rithm and the goodness (e.g. accuracy, distribution) of
the surface points.
It is worth to consider an another aspect in the context
of quality control; it is related to the quality of the prob-
lem statement, or to the question asked. Suppose the
problem is to locate the boundaries of a building in a
laser data set. To sketch a simple case assume the build-
ing has a flat roof and a flat, non-vegetated surround-
ing. Now, grouping the laser points into building top
and ground is a piece of cake. But where exactly is the
boundary? It can only be determined within an uncer-
tainty range that mainly depends on the point spacing
and the size of the building. Hence, the problem as
stated is ill-posed. The fact that the boundary cannot
be precisely located is not a quality problem of the laser
points but a quality problem of the question asked.
In summary we conclude that the quality of derived sur-
face properties depends on the problem stated, the al-
gorithm used to solve it, and on the quality of the raw
laser points, for example their point accuracy and spa-
tial distribution. We elucidate some of the quality control
issues in the next section with experiments on real data.
3 Comparison of Surfaces Obtained from Laser
Ranging and Photogrammetry
The following experiments used data from the Ocean
City test site. Before describing the experiments we
briefly summarize the specifications of the test data.
3.1 Test Site Ocean City
A multisensor data set has been collected over Ocean
City, Maryland, under the auspices of ISPRS WG III/5, the
Geomatics Laboratory for Ice Dynamics of the Byrd Polar
Research Center, and the Photogrammetry Laboratory of
the Department of Civil and Environmental Engineering,
OSU. The data set comprises aerial photography, laser
scanning data, and multispectral and hyperspectral data.
Csathó et al. (1998) provide a detailed description.
Also, the WEB site http://wwwphoto.eng.ohio-state.edu
informs about the current status.
For the experiments we use an aerial stereopair (origi-
nal negatives and digital images, scanned with 28 um
pixel size) and laser scanning data. Large scale aerial
photographs were flown by the National Geodetic Survey
(NGS) at a flying height of 372 m (photo scale approx 1 :
2,435). One strip was triangulated in the classic way, us-
ing GPS ground control points. NASA Wallops made sev-
eral laser data sets available, using the Airborne Topo-
graphic Mapper (ATM) laser system. The ATM is a conical
scanner, developed by NASA for the purpose of measur-
ing ice sheet surfaces. Recently, other applications have
been pursued, for example beach mapping.
The exterior orientation of the photographs is in the
same reference frame as the laser points. Consequently,
features derived from both data sets can be compared
directly.
The photogrammetry laboratory performed an aerial tri-
angulation and several manual measurements. A skilled
operator measured a dense DEM on the Zeiss C120 ana-
lytical plotter. The grid spacing of 2 m compares approx-
imately to the average density of the laser points. In ad-
dition to the DEM, some building outlines and roof tops
have been digitized for comparison with the extracted
features from the laser points. Fig. 3 shows the study
site. We focus on the apartment building in the left part
of the figure and the residential area next to it.
Figure 3: The study site for the experiments reported in this
paper is a small area of the Ocean City Test Site, established by
ISPRS WGIII/5. The apartment complex (left) has a flat roof but
a fairly complicated roof outline. A DEM with 2 m grid spacing
was measured on and around the building. Similarly, a DEM
was measured in the highlighted part of the residential area.
