. Istanbul 2004
the model this
point 2002 the Adjustment of Airborne Laser Altimetry Strips
Sagi Filin 7 George Vosselman
Faculty of Aerospace Engineering
Delft University of Technology, The Netherlands
filinGtechnion.ac.il; M.G.VosselmanQGLR. TUDelft. NL
Commision 3, Working Group 3
f the model is
a, J., Miranda,
the Geomobil
basic relations
1sing, 54; 199-
nt of LIDAR
canning, “3-D
InSAR data”,
)licaciones del
, Barcelona.
Mathematical
ón. A., 2004a.
1 the ICC. 4th
z Technology
CW.. 2004b.
ith GPS/IMU
, Part B3.
ser altimetry
ges 935-942.
. We want to
ny Santiago &
el Soro, Marta
Hernández.
KEY WORDS: Laser altimetry, Error recovery, Strip adjustment, Segmentation
ABSTRACT
Whereas the objectives of a strip adjustment procedure are simple to define, namely improving the accuracy of laser data
and creating a seamless dataset, achieving them is difficult. Complex data acquisition systems and data that carry only little
information are two aspects that make the formulation of a strip adjustment model difficult. Furthermore, difficulties in
processing the data manually and the existence of partial data that usually contain only the laser points but not the system
measurements, are other aspects that should be handled by a strip adjustment algorithm to make it useful. This paper presents
a 3D system-driven strip adjustment algorithm. Based on the properties of the data the error recovery model is surface based.
To eliminate manual processing of the data, handling tie and control information is autonomous, and to have the model
applicable, the input data are the laser points. The application of the procedure is demonstrated here for the estimation of
GPS biases to analyze the magnitude of positional offsets in the data. Results show the existence of significant errors in
position.
1 Introduction
Adjustment of airborne laser scanning data has received grow-
ing attention in recent years. The aim of improving the ac-
curacy of the data and the existence of offsets in the over-
lapping areas of the laser swaths has turned wider attention
to the need for adjustment. The effect of the offsets is not
limited to degraded accuracy but pose great difficulties in
forming a seamless dataset out of the individual laser strips
and complicate significantly further processing of the data.
Indeed, there is still a debate whether laboratory and in-flight
calibration can eliminate those errors, but reality shows that
noticeable errors still exist in the laser data. Their removal
requires the derivation of adjustment procedures for airborne
laser scanning data.
The elimination of the errors from the laser data presents sev-
eral challenges, among which the error model is the central
one. Errors in the systems can be attributed to the individual
components of the data acquisition system (GPS, INS, and
range-finder systems) as well as to their integration; some er-
rors might be constant for a whole mission while others may
vary over space and time. A thorough error modeling does
not guarantee yet their estimation as some errors may have
similar effect as others and result in rank-deficient matrices
or in weak estimation of the parameters. The error model
is therefore not limited to the error analysis but should be
followed by a recoverability analysis of the modeled errors.
In addition, the nature of laser data requires the develop-
ment of adequate algorithms to recover the systematic er-
rors. In contrast to traditional reflectance data that are used
in photogrammetry laser points sample the shape of the over-
flown surface. Point correspondence is practically impossible
to establish under such conditions, and therefore shape based
rather than traditional point based algorithms are needed.
While there is a growing interest in the elimination of sys-
tematic errors from the laser strips the work that has been
*Currently at the Department of Transportation and Geo-information,
Faculty of Civil and Environmental Engineering, Technion-Israel Institute of
Technology.
carried out so far is rather limited. The majority of the re-
ported algorithms consider the offsets only along the height
direction and can be termed as one-dimensional adjustment
procedures (Vaughn et al., 1996; Ridgway et al., 1997; Crom-
baghs et al., 2000; Kager and Kraus, 2001; Kornus and Ruiz,
2003). The level of complexity of the models vary among im-
plementations but their concern with minimizing the height
offsets while ignoring in large positional offsets, is common.
Reference features (control or tie objects) are usually cho-
sen over flat horizontal surfaces where the height offsets are
noticeable, well defined, and easy to measure. Existence of
planimetric offsets in the data (see e.g., Bretar et al., 2003)
will hardly be compensated for with this adjustment and are
likely to remain. Another noticeable shortcoming of com-
mon adjustment procedures is the error recovery model that
is usually chosen. In most cases, the applied procedure mod-
els the biases as transformation parameters on the laser point
coordinates; they can therefore be regarded as data-driven so-
lutions. As was demonstrated in Schenk (2001); Filin (2003a)
some of the errors cannot be compensated for this way and
some offsets will be left in the data untreated. An alter-
native solution in which the system errors are modeled and
solved for allows for the actual errors to be modeled properly
and therefore eliminated. However, as a system driven solu-
tion it requires the system observations as an input (see e.g.,
Burman, 2002). These observations are usually not available
to the end-user so with no access to them this procedure is
rather limited in its applicability.
An optimal solution for the adjustment should opt at model-
ing the actual errors in the system but at the same time be
practical and assume the existence of the type of data that is
usually available to the user. The work presented here con-
cerns the implementation of a system-driven strip adjustment
model. The strip adjustment model is three-dimensional and
accounts for positional as well as height offsets. The im-
plementation assumes that only laser points are given as an
input. Analysis of the model and implementation concerns
are discussed in greater details on Filin (2003b). To assess
the significance of a 3D strip adjustment model the paper an-
Ro emis