ACCURATE REGISTRATION OF ALS DATA WITHOUT CONTROL POINTS
Robert Päquet
School of Engineering, University of Newcastle
Callaghan, NSW 2308, Australia
(robert.paquet@newcastle.edu.au)
KEY WORDS: registration, algorithms, laser scanning, GPS, DEM/DTM, accuracy, matching, adjustment
ABSTRACT:
This paper demonstrates the use of a matching algorithm to register an ALS data set to a reference surface composed
of ground points obtained using kinematic stop-and-go GPS. The matching algorithm minimises the normal distances
between the points of the ALS data set and the facetted reference surface. A particular feature of this experiment is to
sample the reference surface in several high density patches, rather than spreading the sampling evenly over the surface.
The resulting triangulated reference surface is composed of several patches of small triangles in an environment of ex-
tremely large triangles. The weighting technique adopted was based on interpolation errors, which are estimated using
autocorrelation theory. A spatial covariogram is generated and the interpolation errors are determined using an exponen-
tial function fitted on the experimental covariogram. In practice, the accurate registration of ALS data is important, as it
provides the foundation of DEMs used in sensitive areas such as flood studies. The method presented has proven to be a
very successful tool and is an improvement on existing methods.
1. INTRODUCTION
The registration of data sets in the same coordinate sys-
tem is essential for the fusion of data obtained from simi-
lar or different sources. The determination of the registra-
tion accuracy by commercial firms is often undertaken by
comparing points of one surface to heights interpolated be-
tween points on the other surface. The last few years have
seen research undertaken in the development of surface
matching algorithms to automate the registration. High re-
dundancy is achieved with these algorithms, as each point
of one surface can potentially participate in the formation
of a normal equation for a least squares adjustment. These
algorithms should become an important tool for data fu-
sion, especially with data acquisition techniques such as
airborne laser scanning (ALS), which lack thematic infor-
mation.
This paper presents a weighted least squares surface match-
ing algorithm developed at the University of Newcastle,
Australia. The weighting technique used in the algorithm
permits a surface to be registered accurately to a reference
surface with that reference surface covering 20% of the
other surface only.
The accuracy of the registration is related to the density of
the reference data (referred to as S, in this paper) and, to a
lesser degree, to the roughness ofthe surface. A distinction
is made between the precision of the matching, defined by
the residuals of the least squares, and the accuracy of the
registration, which compares the matched points of the sec-
ond surface to their true position. The second surface is the
data set which is transformed by the least squares match-
ing algorithm in the coordinate system of Sy: it is referred
to as 55 in this paper.
The aim ofthis paper is to present the weighted least squares
software and to demonstrate its performance through an
application. The application involves registering a set 59
of 27,000 points in the coordinate system of 9,, a set of
900 points. S; is a patchy set obtained with stop-and-go
global positioning system method (GPS), while 5» is a set
of points filtered from a larger set sampled with an airborne
laser scanner.
2. THE TOOLS OF THE EXPERIMENT
2.1 Least squares Matching Algorithm
2.1.1 Background: Surface matching, also referred to
as registration without control points, describes an auto-
mated method used to find the parameters of a transfor-
mation which minimises the separation between two 2.5D
surfaces. Early matching algorithms were presented by
Rosenholm and Torlegärd (1988) and Ebner and Strunz
(1988). In both instances, better results were reported us-
ing a gridded DEM to orient photogrammetric data than
with conventional methods. Similar methods are used in
deformation studies and monitoring (Pilgrim, 1996). Reg-
istration methods using a control surface are justified par-
ticularly in areas where permanent control markers are not
possible (abdominal deformation in pregnancy (Karras and
Petsa, 1993)), unethical (dental erosion (Mitchell, 1995;
Mitchell and Chadwick, 1998, 1999)) or impractical (coastal
erosion (Buckley, 2003; Mills et al., 2003)). Maas (2002)
developed a least squares matching (LSM) technique for
laser scanner data strips adjustment: systematic deforma-
tions due to GPS/INS systems are corrected by solving for
the three shift parameters T'x, Ty and Tz.
The algorithms mentioned minimise the difference in heights
between the points of one surface to the facets of the other
surface. The algorithm presented in this research min-
imises the normal distances between the points of 55 t0
1172
the
wh
[s
wit
An
by
kee
whe
Dy
para
Aw,
to ir
Let
dist:
Equ:
With
Equa:
