Hans-Gerd Maas
Least-Squares Matching with Airborne Laserscanning Data
in a TIN Structure
Hans-Gerd Maas
Institute of Photogrammetry and Remote Sensing
Dresden Technical University
Helmholtzstr. 10
D-01062 Dresden, Germany
e-mail hmaas@rcs1.urz.tu-dresden.de
Keywords:
Airborne laserscanning, least-squares matching, TIN, accuracy
Abstract:
A number of tasks in airborne laserscanning require the establishment of correspondences between point data from
neighbouring strips, or referencing between point clouds and object models. These tasks may be solved by interpolating
laserscanner data, which are usually irregularly distributed 2'/,-D points, to a regular grid and applying standard photo-
grammetric matching techniques. Instead, the paper presents a formulation of least squares matching based on the
original data points in a triangulated irregular network structure, thus avoiding degrading effects caused by the interpo-
lation. The technique determines shifts in all three coordinate directions together with their covariance matrix. It can be
shown that applying matching techniques to laserscanner data causes large systematic errors of the shift parameters in
the case of partial occlusions. The presented formulation on the basis of a TIN structure allows for manifold extensions
to solve this problem.
The technique and a number of extensions have been implemented and applied to the measurement of strip errors in an
airborne laser scanner dataset with moderate point density, consisting of 20 strips including crossing strips. The paper
shows the results from this test, discusses the advantages of the presented technique and the limitations of matching
techniques applied to laserscanner data. Special attention has to be paid to problems caused by height discontinuities in
the data and by the fact that the design matrix in least squares matching is derived from observations with stochastic
properties. The latter leads to precision figures that are usually much too optimistic. A detailed analysis of the design
matrix and extensive testing lead to better funded precision figures for the standard deviation of the obtained shift
parameters. These are in the order of one centimeter in height direction and one decimeter in horizontal direction,
corresponding to about '/;sth of the average point spacing.
1. Introduction
Least-squares matching (LSM) is a technique that is being applied regularly by photogrammetrists for the establishment
of correspondences between images taken from different viewing points, or between subsequent images of an image
sequence. Formulated for two-dimensional greyscale images (F ‘rstner 1984, Grhn 1985), it determines the parameters of
an affine transformation between corresponding patches of two or more images. Typical application fields of LSM are the
determination of homologous points between consecutive images and image strips in a conventional aerotriangulation,
or the matching between strips of linear array cameras.
Just like conventional photogrammetric image data, airborne laserscanning data of larger areas is also acquired in a
stripwise manner. Basically, the 2'/,-D point clouds generated by airborne laserscanning are directly geo-referenced due
to the use of GPS/INS systems onboard the aircraft. Because of errors of these instruments or sub-optimalities of the
GPS/INS integration, however, points of neighbouring laserscanner strips will usually show vertical discrepancies in the
order of several centimeters and horizontal discrepancies in the order of a few decimeters (Huising and Gomes Pereira,
1998). The significance of these discrepancies depends on the application and the point density: in datasets with a point
density of approximately one point per ten square metres, as are being used for the generation of digital elevation
models, one will often only notice the vertical discrepancies; in datasets with several points per square meter used for
the generation of 3-D city models, both vertical and horizontal discrepancies are often clearly noticeable.
548 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
Sever
techn
matcl
on the
strip :
two té
Planii
to a ri
interp
uted 1
In the
and f
probk
In che
2. B
The b
of LS
param
techn;
are di
2.1 ]
The b
discre
out of
D dat
coord
the si
minin
are nc
match
in on
corres
Obser
origin
for tl
subtra
interp
corres
other
of the
norme
Form
assigr
proce
2.2 ]
The k
deterr
for m
soluti
where