Association-Matrix-Based Sample Consensus
Approach for Automated Registration of
Terrestrial Laser Scans Using Linear Features
Kaleel Al-Durgham and Ayman Habib
Abstract
This paper presents an approach for the automatic registra-
tion of terrestrial laser scans using linear features. The main
contribution here is introducing a new matching strategy that
uses an association matrix to store information about candi-
date matches of linear features. The motivation for this work
is aiding the 3
D
modeling of industrial sites rich with pole-like
features. The proposed matching strategy aims at establish-
ing hypotheses about potential minimal matches of linear
features that could be used for the estimation of the transfor-
mation parameters relating the scans; then, quantifying the
agreement between the scans using the estimated transforma-
tion parameters. We combine the association matrix and the
well-known
RANSAC
approach for the derivation of conjugate
pairs among the two scans. Rather than randomly selecting
the line pairs as in the
RANSAC
-based registration, the asso-
ciation matrix guides the process of selecting the candidate
matches of linear features. Experiments are conducted using
laser scanning data of an electrical substation to assess
the performance of the proposed association-matrix-based
sample consensus approach as it compares to the tradition-
al
RANSAC
-based procedure. The association-matrix-based
approach showed consistent tendency of bringing up the
correct matches first before the
RANSAC
-based registration.
Introduction
The ability of collecting dense 3
D
spatial data without the
need for the direct contact together with the low operational
cost with few crew members encouraged institutions and
individuals to utilize terrestrial laser scanning systems in
wide range of applications such as digital building model
generation, industrial site modeling, cultural heritage docu-
mentation, and other civilian and military applications. The
acquisition of several laser scans with significant overlap is
one of the fundamental requirements to guarantee full
coverage of the site of interest and to attain more details than
what could be achieved from a single scan. The raw outcome
of a single terrestrial scan is a cloud of 3
D
coordinates that are
defined with respect to a local coordinate system associated
with the scanner’s location and orientation. Hence, a registra-
tion process has to be performed when dealing with multiple
scans in order to align these scans to a common reference
frame. A registration process aims at estimating the
3
D
-Helmert transformation parameters that describe the
absolute orientation parameters between the involved scans.
The 3
D
-Helmert transformation parameters include three
rotation angles, three translations, and a scale factor. For a
well-calibrated laser scanner, the scale factor is considered
unity since the laser ranging principle provides the true scale.
To date, a vast amount of research work with various method-
ologies has been conducted on the registration of laser scans
(Liang
et al
., 2014; Theiler and Schindler, 2012; Yao
et al
.,
2010). According to Habib and Al-Ruzouq (2004), a registra-
tion process should address four issues: (a) The registration
primitives which are the conjugate features within the scans
that can be used to solve for the transformation parameters.
These primitives could be points, 3
D
lines, and/or planes; (b)
The transformation parameters relating the reference frames of
the involved datasets; (c) The similarity measure, which is the
mathematical constraint that describes the coincidence of
conjugate primitives after the registration process; and (d) The
matching strategy, which represents the guiding framework for
the automatic registration process. Registration algorithms can
be divided into three main categories: (a) point-based algo-
rithms that use points as the registration primitives; (b)
feature-based algorithms that use other geometric features as
the registration primitives; and (c) direct georeferencing
algorithms that incorporate additional sensors such as
GNSS
and
INS
with the laser scanner. The point-based algorithms are
usually used to establish the fine alignment between the
overlapping scans, while the feature-based and the direct
georeferencing registration algorithms target the problem of
establishing the coarse alignment between the scans. The
well-known Iterative Closest Point (
ICP
) (Besl and McKay,
1992; Chen and Medioni, 1992; Zhang, 1994) is an example of
the most commonly used point-based registration algorithm,
which minimizes the point-to-point distance in the overlap-
ping area between the different scans. The Iterative Closest
Patch (
ICPatch
) (Habib
et al
., 2010) is a variant of the
ICP
where
points in one scan and a triangular irregular network in
another scan serve as the registration primitives. The transfor-
mation parameters are estimated by minimizing the sum of the
squared normal distances between the conjugate point-patch
pairs. Another derivative of the
ICP
is the Iterative Closest
Projected Point (
ICPP
), which aims at minimizing the distance
between a point in one scan and its projection on the plane
defined by the closest three points in the other scan (Al-
Durgham and Habib, 2013). Point-based registration algo-
rithms were proven to be accurate. However, in order to
establish the primitives correspondence (i.e., point to its
nearest point in another scan or point to its nearest patch), a
coarse alignment of the scans being registered is required. In
Kaleel Al-Durgham is with the Digital Photogrammetry Re-
search Group (DPRG), Department of Geomatics Engineering
Schulich School of Engineering, University of Calgary (kmal-
).
Ayman Habib is with Lyles School of Civil Engineering, Pur-
due University (
).
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 11, November 2014, pp. 1029–1039.
0099-1112/14/8011–1029
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.11.1029
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
November 2014
1029
