CA, 9-11 Nov. 1999
ig imagery to provide edge
ited by several researchers
98; Csathó et al., 1999; Haala
nain concept of the approach
paper utilizes the beneficial
c data and laser data to produce
imagery can provide accurate
ig the location of surface
provides accurate elevations,
te location information of the
ted in Figure 1. Thus the two
n the advantages of each, and
1 of an accurate surface model.
has been undertaken using an
faryland. Laser data and aerial
> day by NASA and NGS
inary experiments have been
gorithm. This paper presents
ia fusion components, describes
e results from the initial
ATION APPROACH
aper utilizes information from
ammetric data to produce an
face. The conceptual approach
st phase, surfaces created from
stered to the same coordinate
s of extracting edge information
hich is used to delineate surface
ene. By merging the edge
surface generation, a DSM that
scene can be produced.
ndertaken to determine the
en the laser surface and the
ice. Theoretically, the two data
ordinate system, however the
e laser data may introduce à
o surfaces, which must be
may be performed accurately
nisalignment has been observed
tial testing of the algorithm,
surface registration component
to find the transformation
These parameters are used to
he coordinate system of the
data have higher planimetric
data (Baltsavias, 1999). The
neters must be as accurate as
ary degradation occurs in the
from merging the data sets.
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999
The algorithm was developed specifically for matching visible
surface models of urban areas produced using different
acquisition methods, and in particular, surfaces from airborne
laser scanner data and surfaces automatically generated using
photogrammetric data. The registration algorithm solves for three
translation parameters, three rotation parameters and one scale
factor. The surface points are not interpolated to a regular grid
and there are no identifiable conjugate points in the surfaces. For
more detailed information on the algorithm and the results of
testing, the reader is referred to Postolov et al. (1999).
2.2 Data Fusion
The main emphasis of the research project is to develop an
algorithm to produce accurate visible surface models of urban
areas, utilizing both laser data and photogrammetric data.
The data fusion component utilizes edges extracted from the
stereo imagery to obtain accurate horizontal locations of surface
discontinuities. The edges are defined in three dimensions and
are used as breaklines when merged with the transformed laser
data. A new surface is generated using the merged data, which is
expected to have a higher accuracy than the surface derived from
either of the separate data sets.
The conceptual approach formulated in this research proposes
that the edges representing surface discontinuities are
automatically derived as 3D line segments from the stereo
imagery using edge extraction and feature matching techniques.
Initial research has been undertaken into the automatic extraction
of 3D line segments.
The edge pixels are detected in each image of the stereo pair of
digital images. At present, the optimal zero-crossing operator
(Sarkar and Boyer, 1991) is used for the detection of the edge
pixels. The edge pixels are analyzed to find connected lines and
using these lines, straight line segments are determined. Line
segments are used as they adequately describe man-made objects
and can be more accurately located than point features when
using feature matching techniques (Fradkin and Ethrog, 1997).
Also, each line segment can be easily described using the start
and end points of the segment and the gradient of the line.
For each straight line segment in the first image, the second
image is searched for a matching segment. To make the
searching of the images more computationally efficient, a coarse
to fine approach is utilized, and constraints are applied to the
search process, including epipolar and ordering constraints
(Grewe and Kak, 1994). The stereo imagery has exterior
orientation parameters which allow the 3D coordinates of the line
segments to be determined when the matching of the line
segments is completed.
The automatic extraction of edges has certain limitations which
must be addressed. Using automatic extraction methods, not all
surface discontinuities are detected due to factors such as
foreshortening and lack of contrast in the imagery. There might
be a number of incomplete or incorrect breaklines, and other
breaklines may be detected that do not represent surface
discontinuities may be detected. Instead, these breaklines may
refer to visual edges in the images, which are purely changes in
grey value, such as line markings on roads. The effect of
including these breaklines in the data fusion process is to be
investigated.
For this research, the most important edges to be detected are the
discontinuities between the roof and the ground. These areas are
important to the accurate representation of the visible surface, as
they contain dramatic changes in elevation.
The detected breaklines are expected to be mainly rooflines,
however to properly define the surface, the ground level near the
rooflines must also be determined. Therefore, the roofline will be
projected vertically onto the ground surface to produce a new
breakline, thus describing the surface at ground level as well as
roof height. The location of the new breakline must be slightly
offset outward from the roofline, as the triangulation process
being used does not allow points to exist with the same horizontal
location, as would be the case for a vertical plane.
The elevation of the ground surface at the location of the foot of
the building must be determined. It is proposed that the laser
data in the areas surrounding the breakline be searched. In the
general case of the breakline delineating a surface discontinuity
which is a roofline, it is expected that one side of the breakline
will have points of higher elevation than the other side. The
lower elevations will be assumed to be the elevation at ground
level. Therefore, the new breakline will be generated and
assigned the elevation of the ground points in that area.
In the current testing of the data fusion algorithm, manually
measured breaklines are used. These breaklines are adequate to
test the validity of the algorithm and to assess the implementation
of the approach.
The laser points which occur on the breaklines are eliminated,
again due to the triangulation process not accepting points with
identical horizontal locations. The triangulation process is used
to generate a TIN which utilizes the laser data points and
constrains the triangles to follow the breaklines. The inclusion of
the elevation information to the triangulation provides the digital
surface model produced using the laser data and the
photogrammetric data.
4 PRELIMINARY TESTING AND RESULTS
The initial testing was carried out using a data set over Ocean
City, MD, which includes digital stereo imagery and laser data.
The data set covers different types of areas, including residential
areas, flat terrain, beach front, dunes, canals and high-rise
buildings. Only a small portion of a residential area has been
used for the initial testing of the algorithm.
The laser data have been filtered so as to only retain the first laser
measurement and not any subsequent measurements from the
laser sensor. The laser data has also been transformed to the
coordinate system of the photogrammetric data using the
parameters determined by the process surface registration. The
two data sets are therefore on the same coordinate system and are
representing the same surface. Laser points in the new data set
are eliminated if they occur within close proximity to the
breaklines. A filtering process is undertaken to eliminate points
within a horizontal buffer zone of the line segments representing