AUTOMATED REGISTRATION OF UNORGANISED POINT CLOUDS FROM TERRESTRIAL
LASER SCANNERS
Kwang-Ho Bae and Derek D. Lichti
Department of Spatial Sciences, Curtin University of Technology, GPO Box U1987, Perth, WA, 6845, Australia
baek@vesta.curtin.edu.au, d.lichti@curtin.edu.au
Commission V, WG V/2
KEY WORDS: Registration, Automation, Three-dimensional, Point Cloud, Laser Scanning, TLS
ABSTRACT
Terrestrial laser scanners provide a three-dimensional sampled representation of the surfaces of objects resulting in a very
large number of points. The spatial resolution of the data is much higher than that of conventional surveying methods.
Since laser scanners have a limited field of view, in order to obtain a complete representation of an object it is necessary
to collect data from several different locations that must be transformed into a common coordinate system. Existing
registration methods, such as the Iterative Closest Point (ICP) or Chen and Medioni's method, work well if good a priori
alignment is provided. However, in the case of the registration of partially overlapping and unorganised point clouds
without good initial alignment, these methods are not appropriate since it become very difficult to find correspondence.
A method based on geometric primitives and neighbourhood search is proposed. The change of geometric curvature and
approximate normal vector of the surface formed by a point and its neighbourhood are used to determine the possible
correspondence of point clouds. Our method is tested with a simulated point cloud with various levels of noise and two
real point clouds.
1 INTRODUCTION
The registration of point clouds is one of the important
steps in processing data from laser scanners. Excellent
reviews about the existing methods can be found in Har-
alick et al. (1989) and Rusinkiewicz and Levoy (2001).
To the best of the authors’ knowledge, a method for the
registration of partially overlapping point clouds from ter-
restrial laser scanners without good a priori alignment has
not developed yet. The most significant reason is that it is
very difficult to recover the correspondence between point
clouds without good a priori alignment. In this paper a
method for the registration of two partially overlapping
point clouds measured from different locations using geo-
metric primitives and neighbourhood search without good
a priori alignment is proposed. Sharp et al. (2002) pro-
posed a method based on Euclidean invariant features: cur-
vature, second order moments, and spherical harmonics.
Our method is based on the change of curvature rather
than curvature and is a thresholding method such as Zhang
(1994).
The Iterative Closest Point (ICP) algorithm is a reliable and
popular method for point cloud registration (Horn, 1987,
Besl and McKay, 1992). Horn developed an adjustment
method for recovery of unknown transformation parame-
ters with knowledge of the correspondence of point clouds.
Besl and McKay developed the ICP using Horn's algo-
rithm in conjunction with a neighbourhood search algo-
rithm. If approximate a priori information about the point-
to-point correspondence of two point clouds is provided,
then the ICP can iteratively recover the rigidbody transfor-
mation that aligns two point clouds. It converges mono-
tonically to a local minimum, which may or may not be
the global minimum, and all points in a point cloud are
assumed to have correspondence in the other cloud.
Suppose that there are two point clouds, C. and C?, and
they are measured from different locations. A point cloud
C! hasa set of Nc: points, (pl, -- Pla }. Bold and nor-
mal letters represent a vector and a scalar, respectively. Let
Ip} — p3|| be the distance between point p; of point cloud
C! and p? of C?. Let C P(p], C?) be the corresponding
point in C? ofa point p]. The ICP algorithm can be briefly
described as follows.
I. Assume that the point in C? closest to a point in C? .
is the corresponding point.
N
Find the correspondence of two point clouds,
Ne ip v
C= en ence MC P(Tuersk (pl). 2},
where C is the set of all pairs of corresponding
points, Tizer=k is the transformation of the kth
iteration and 77,,,.—9 is an initial transformation. C
may or may not be one-to-one matching.
3. Compute the new transformation T{rer=r+1 that
minimizes the sum of square distances between
corresponding point pairs, i.e.
Miter=k
s Ip! ET CPG sp cog
i=]
where 72;ter=z is the number of sample in Ath
iteration.
Since the ICP is based only on a local search algorithm
to recover correspondence between two point clouds and
it minimises the sum of square distances between possible
corresponding points, it converges slowly sometimes and
tends to fall into local minima.
Internatior
Another :
to-surface
gorithm (
of the squ
face. Thi:
ever, the
each othe
The ICP,
assume tl
estimate :
point clo
ori geore:
rect. Al
other me:
is not alv
good reg
they only
algorithir
tion, whi
closed-fo
tion of ir
as conver
2 PRO
Geometri
vature, ar
additiona
dence of
dence of
a local se
and the «
sional rig
tated acc
or Chen ¢
mal vectc
vature of
ria for se
The angl
p; can be
where n,
vectors 0
ture betw
where M
of curvat
estimatec
bourhooc
as the rat
2.1 Noi
The norn
eigenvec
