The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
1148
The objective of this paper is to present a comparative analysis
of indirect georeferencing using real data to evaluate the use of
three sources of control data, namely, ground control points,
LiDAR patches, and LiDAR lines. Section 2 of this paper
describes indirect georeferencing using LiDAR patches and
LiDAR lines. A semi-automated approach for the extraction of
patches and lines from LiDAR data is presented and used to
obtain the image EOP from LiDAR control features. In Section
3, quantitative analysis, using RMSE analysis, and qualitative
analysis, using orthoimage generation, are presented to
demonstrate the comparison between the object reconstruction
results using the three sources of control data. Finally, the
conclusions are summarized in Section 4.
2. GEOREFERENCING METHODS
Photogrammetric georeferencing is the process of relating the
image and ground coordinate systems by defining the position
and orientation information (EOP) of the camera at the moment
of exposure relative to the object space coordinate system.
Computing the EOP is performed either directly (direct
georeferencing) when GPS/INS is available onboard the
imaging platform, or indirectly (indirect georeferencing) using
ground control information. Ground control points (GCPs) are
traditionally the most commonly used source of control for
photogrammetric indirect georeferencing. Despite its proven
accuracy in photogrammetric reconstruction, using GCPs as a
source of control is costly and labour intensive as it requires
field surveying. For this reason, other sources of control data
require investigation. The availability of LiDAR data, however,
allows for alternative methods for image georeferencing. In
particular, LiDAR-derived control features can be utilized for
the georeferencing of the photogrammetric data relative to the
LiDAR reference frame. Since the LiDAR footprints are
irregularly distributed, however, no point-to-point
correspondence can be assumed between the photogrammetric
and LiDAR data. As such, it is almost impossible to identify
distinct conjugate points in overlapping photogrammetric and
LiDAR data. Consequently, LiDAR patches (Section 2.1) and
LiDAR lines (Section 2.2) will be used as control information
for the georeferencing of the photogrammetric data.
2.1 Indirect Georeferencing using LiDAR patches
This section outlines the procedure in which LiDAR patches are
used to obtain the image georeferencing parameters. The steps
for this procedure include the extraction of the LiDAR patches
from the LiDAR point cloud, followed by the incorporation of
LiDAR-derived areal features in the photogrammetric
triangulation procedure.
2.1.1 Extraction of LiDAR Patches: In order to utilize
LiDAR control features, the features must be extracted from the
LiDAR point cloud. This section outlines the process for the
extraction of areal features from irregular LiDAR footprints
(Al-Durgham, 2007). The extraction of features from LiDAR is
performed using a developed program. The process begins by
displaying the LiDAR intensity images in the program window,
in which the user selects an area where it appears that areal
features might be present. The user clicks on the centre of the
area after defining the radius of a circle within which the
original LiDAR footprints will be extracted. It should be noted
that the LiDAR intensity images are only used for visualization
purposes. Figure la shows a sample area as well as the original
LiDAR footprints located in a selected area. Then, a
segmentation technique (Kim et al., 2007) is used to identify
planar patches in the point cloud within the selected area. The
outcome from the segmentation is an aggregated set of points
representing planar patches in the selected area (Figure lb).
(a) (b)
Figure 1: a) Area of interest selection and LiDAR point cloud
extraction, b) Segmented planar patches.
2.1.2 Incorporation of Areal Features for Image
Georeferencing: The approaches used to incorporate areal
features extracted from LiDAR data in a photogrammetric
triangulation procedure are now presented. The first outlined
approach is the coplanarity-based incorporation of areal
features. The second approach involves the use of a point-based
incorporation of areal features, where the weight matrix of the
utilized points is modified.
Coplanarity-based Incorporation of Planar Patches
In this approach, the planar patch in the imagery is defined by a
minimum of three points, for example points a, b, and c, which
are located in the image space, while the LiDAR patch is
defined as a set of LiDAR points in object space (Habib et al.,
2007). The points, a, b, and c should be visible in at least two
overlapping images. The collinearity equations are used to
relate the image space coordinates of the points a,b,c to their
object space coordinates, A,B,C. The LiDAR points belonging
to a certain planar-surface patch should coincide with the
photogrammetric patch representing the same object space
surface. The coplanarity of the LiDAR and photogrammetric
points is explained mathematically in Habib et al. (2007). In
physical terms, this constraint means that the normal distance
between any LiDAR point P and the corresponding
photogrammetric surface consisting of the three points should
be zero. In other words, the volume of the tetrahedron
comprised of the four points (A, B, C and P) should be equal to
zero, as these points belong to the same plane. This constraint is
applied for all LiDAR points located within this surface patch.
Point-based Incorporation of Planar Patches
A second approach for the incorporation of planar patches uses
a point-based technique, in which existing bundle adjustment
procedures which are based on the collinearity equations, can
be used for the incorporation of control areal features
(Aldelgawy et al., 2008). For this approach, conjugate patch
vertices are defined in at least two overlapping images. Then,
the corresponding control patch is extracted from the LiDAR
point cloud using the procedure described in section 2.1.1.
From the extracted LiDAR control patch, which consists of
hundred of points, some points are selected as patch vertices.
The number of points selected in the LiDAR patch should be
equivalent to the number of vertices defined in the imagery.