Kerry McIntosh
errors introduce a misalignment between these two surfaces. This misalignment must be eliminated before data fusion
may be performed accurately (Kraus and Pfeifer, 1998).
Several factors were considered when designing the matching algorithm, including the fact that no conjugate points
occur in the data sets, and that the data sets may have different spatial frequencies. The transformation parameters are
used to transform the laser data to the photogrammetric coordinate system, as the horizontal accuracy of the
photogrammetric data set is more easily verified using ground control and visually identifiable points.
The second phase of the proposed approach combines information from each data source to obtain the optimal
topographic surface. The photogrammetric data is utilized by extracting edges from the stereo imagery, and processing
the edges using feature-based matching techniques. These edges are used to obtain accurate locations of the surface
discontinuities in the urban scene. The edges are defined in three dimensions and are used as breaklines when merged
with the laser data. The laser data is filtered to eliminate points that are close to the newly imported breaklines. This
reduces the probability of erroneous points being included in the surface, as random errors in the laser data such as
corner reflections will occur close to these areas. A new surface is generated using the merged data. This surface is
expected to have a higher accuracy than either surface derived from the separate data sets.
The approach presented in this paper utilizes the beneficial properties of both photogrammetric data and laser data to
produce an accurate DSM. Testing of the research approach has been undertaken using an urban site covering Ocean
City, Maryland, USA. Laser data and aerial images, acquired on the same day by NASA and NGS respectively, are
used. Preliminary experiments have been performed to test and refine the algorithm. This paper presents the surface
registration and the data fusion components, describes the data set and details the results from the initial
experimentation.
2 BACKGROUND INFORMATION
Digital surface models of urban areas may be created using different methods, such as digital photogrammetric
processing or by using airborne laser scanner data. Each of these methods has benefits and limitations.
Digital photogrammetric methods of automatic surface reconstruction have become widely used due to the efficiency
and cost effectiveness of the production process, especially in open or flat areas, and when using small and medium
scale imagery (Krzystek and Ackermann, 1995). However, most software packages perform poorly in areas with abrupt
height differences, such as those occurring frequently in urban areas (Haala, 1999). The degradation in performance
can be caused by failures of the image matching process (Axelsson, 1998). Such failures may be due to factors like lack
of texture in the images (Haala, 1994), poor image quality, shadows in the images, occlusions, surface discontinuities
(Haala ef al., 1997) and foreshortening. The problems occur when using digital photogrammetry in urban areas and
result in inaccuracies in the DSM, which can be seen in the smoothing effect on surface discontinuities (Baltsavias,
1999: Haala, 1999; Toth and Grejner-Brzezinska, 1999).
One of the benefits of photogrammetry is that the imagery contains more information than just the position of pixels in
the images. Gray-value changes in the images allow the identification and classification of objects, such as buildings or
vegetation, and can be used to detect edges in the images, which often indicate the location of surface discontinuities
(Baltsavias, 1999; Fradkin and Ethrog, 1997; Haala and Anders, 1997).
Laser scanning is recognized as an accurate data source for DSM generation in urban areas (Haala ef al., 1997). The
spatial resolution of the data is dependent on several factors, such as flying height, flying speed and scanner frequency
(Lemmens et al, 1997). Characteristics and performance of laser data systems have been discussed by many
researchers (Ackermann, 1999; Axelsson, 1998; Baltsavias, 1999; Fritsch, 1999; Hug and Wehr, 1997; Kilian ef al.,
1996). Calibration methods and the errors that may occur in the data have also been investigated (Fritsch and Kilian,
1994; Huising and Gomes Pereira, 1998; Lemmens ef al, 1997). Data processing and filtering methods have been
described by Axelsson (1999), Hug and Wehr (1997) and Kilian ef al. (1996).
Laser data provides accurate points with high spatial frequency, however breaklines are not present in the data
(Ackermann, 1999; Axelsson, 1999: Haala ef al., 1997; Kraus and Pfeifer, 1998), and therefore the position of surface
discontinuities can only be estimated or calculated by methods such as segmentation of the range data (Haala et al.,
1997). To illustrate this point, Figure 1 presents an elevation image generated from laser data showing high-rise
buildings. The edges of the buildings are not well defined, though a high spatial density of the laser data points is
indicated by the ‘ragged’ nature of the edges, as also noted by Vosselman (1999).
564 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
Inve
pho
con
(Bal
lase
The
beet
reite
199
to p
al.,
al.
The
an a
DSN
3.1
The
phot
how
musi
The
The
in th
The
acqu
using
para
conj
refer
3.2
The
discc
trans
than
prop
stere