Full text available at: http://dx.doi.org/10.

1561/2300000035

A Review of Point Cloud

Registration Algorithms for
Mobile Robotics

François Pomerleau
Autonomous Space Robotics Laboratory
University of Toronto
Toronto, Canada
[email protected]
Francis Colas
Inria, Villers-lès-Nancy, F-54600
CNRS, Loria, UMR 7503, Vandœuvre-lès-Nancy, F-54500
Université de Lorraine, Vandœuvre-lès-Nancy, F-54500
France
[email protected]
Roland Siegwart
Autonomous Systems Lab
ETH Zurich
Zurich, Switzerland
[email protected]

Boston — Delft
 Full text available at: http://dx.doi.org/10.1561/2300000035

Foundations and Trends R in Robotics

Published, sold and distributed by:

now Publishers Inc.
PO Box 1024
Hanover, MA 02339
United States
Tel. +1-781-985-4510
www.nowpublishers.com
[email protected]
Outside North America:
now Publishers Inc.
PO Box 179
2600 AD Delft
The Netherlands
Tel. +31-6-51115274
The preferred citation for this publication is
F. Pomerleau, F. Colas and R. Siegwart. A Review of Point Cloud Registration
Algorithms for Mobile Robotics. Foundations and Trends R in Robotics, vol. 4,
no. 1, pp. 1–110, 2013.
This Foundations and Trends R issue was typeset in LATEX using a class file designed
by Neal Parikh. Printed on acid-free paper.
ISBN: 978-1-68083-025-5
c 2015 F. Pomerleau, F. Colas and R. Siegwart

All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted in any form or by any means, mechanical, photocopying, recording
or otherwise, without prior written permission of the publishers.
Photocopying. In the USA: This journal is registered at the Copyright Clearance Cen-
ter, Inc., 222 Rosewood Drive, Danvers, MA 01923. Authorization to photocopy items for
internal or personal use, or the internal or personal use of specific clients, is granted by
now Publishers Inc for users registered with the Copyright Clearance Center (CCC). The
‘services’ for users can be found on the internet at: www.copyright.com
For those organizations that have been granted a photocopy license, a separate system
of payment has been arranged. Authorization does not extend to other kinds of copy-
ing, such as that for general distribution, for advertising or promotional purposes, for
creating new collective works, or for resale. In the rest of the world: Permission to pho-
tocopy must be obtained from the copyright owner. Please apply to now Publishers Inc.,
PO Box 1024, Hanover, MA 02339, USA; Tel. +1 781 871 0245; www.nowpublishers.com;
[email protected]
now Publishers Inc. has an exclusive license to publish this material worldwide. Permission
to use this content must be obtained from the copyright license holder. Please apply to
now Publishers, PO Box 179, 2600 AD Delft, The Netherlands, www.nowpublishers.com;
e-mail: [email protected]
 Full text available at: http://dx.doi.org/10.1561/2300000035

Foundations and Trends R in Robotics

Volume 4, Issue 1, 2013
Editorial Board

Editors-in-Chief

Henrik Christensen Roland Siegwart

Georgia Institute of Technology ETH Zurich
United States Switzerland

Editors

Minoru Asada Simon Lacroix

Osaka University Local Area Augmentation System
Antonio Bicchi Christian Laugier
University of Pisa INRIA
Aude Billard Steve LaValle
EPFL UIUC
Cynthia Breazeal Yoshihiko Nakamura
MIT University of Tokyo
Oliver Brock Brad Nelson
TU Berlin ETH Zurich
Wolfram Burgard Paul Newman
University of Freiburg Oxford University
Udo Frese Daniela Rus
University of Bremen MIT
Ken Goldberg Giulio Sandini
UC Berkeley University of Genova
Hiroshi Ishiguro Sebastian Thrun
Osaka University Stanford University
Makoto Kaneko Manuela Veloso
Osaka University Carnegie Mellon University
Danica Kragic Markus Vincze
KTH Stockholm Vienna University
Vijay Kumar Alex Zelinsky
University of Pennsylvania CSIRO
 Full text available at: http://dx.doi.org/10.1561/2300000035

Editorial Scope

Topics

Foundations and Trends R in Robotics publishes survey and tutorial

articles in the following topics:

• Mathematical modelling • Software systems and

architectures
• Kinematics
• Sensors and estimation
• Dynamics • Planning and control

• Estimation methods • Human-robot interaction

• Industrial robotics
• Artificial intelligence in
robotics • Service robotics

Information for Librarians

Foundations and Trends R in Robotics, 2013, Volume 4, 4 issues. ISSN

paper version 1935-8253. ISSN online version 1935-8261. Also available
as a combined paper and online subscription.
 Full text available at: http://dx.doi.org/10.1561/2300000035

Foundations and Trends R in Robotics

Vol. 4, No. 1 (2013) 1–104
c 2015 F. Pomerleau, F. Colas and R. Siegwart
DOI: 10.1561/2300000035

A Review of Point Cloud Registration

Algorithms for Mobile Robotics

François Pomerleau
Autonomous Space Robotics Laboratory
University of Toronto
Toronto, Canada
[email protected]
Francis Colas
Inria, Villers-lès-Nancy, F-54600
CNRS, Loria, UMR 7503, Vandœuvre-lès-Nancy, F-54500
Université de Lorraine, Vandœuvre-lès-Nancy, F-54500
France
[email protected]
Roland Siegwart
Autonomous Systems Lab
ETH Zurich
Zurich, Switzerland
[email protected]
 Full text available at: http://dx.doi.org/10.1561/2300000035

Contents

1 Twenty years of ICP: The Legacy 2

1.1 Early Solutions . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Division and Explosion of the Field . . . . . . . . . . . . . 7
1.3 Algorithm Overview . . . . . . . . . . . . . . . . . . . . . 10
1.4 Overview of the Review . . . . . . . . . . . . . . . . . . . 12

2 Formalization of the ICP Solution Family 15

2.1 Reading and Reference Sources . . . . . . . . . . . . . . . 16
2.2 Transformation Functions . . . . . . . . . . . . . . . . . . 20
2.3 Data Filters . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 Association Solver . . . . . . . . . . . . . . . . . . . . . . 32
2.5 Outlier Filters . . . . . . . . . . . . . . . . . . . . . . . . 37
2.6 Error Minimization . . . . . . . . . . . . . . . . . . . . . . 39
2.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3 Registration Use Cases 47

3.1 Search and Rescue . . . . . . . . . . . . . . . . . . . . . . 48
3.2 Power Plant Inspection . . . . . . . . . . . . . . . . . . . 62
3.3 Shoreline Monitoring . . . . . . . . . . . . . . . . . . . . 67
3.4 Autonomous Driving . . . . . . . . . . . . . . . . . . . . . 73
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

ii
 Full text available at: http://dx.doi.org/10.1561/2300000035

iii

4 Conclusion 84

Acknowledgements 90

Appendices 91

A Derivation for Point-to-Plane Error 92

References 95
 Full text available at: http://dx.doi.org/10.1561/2300000035

Abstract

The topic of this review is geometric registration in robotics. Registra-

tion algorithms associate sets of data into a common coordinate system.
They have been used extensively in object reconstruction, inspection,
medical application, and localization of mobile robotics. We focus on
mobile robotics applications in which point clouds are to be registered.
While the underlying principle of those algorithms is simple, many
variations have been proposed for many different applications. In this
review, we give a historical perspective of the registration problem and
show that the plethora of solutions can be organized and differentiated
according to a few elements. Accordingly, we present a formalization
of geometric registration and cast algorithms proposed in the litera-
ture into this framework. Finally, we review a few applications of this
framework in mobile robotics that cover different kinds of platforms,
environments, and tasks. These examples allow us to study the specific
requirements of each use case and the necessary configuration choices
leading to the registration implementation. Ultimately, the objective of
this review is to provide guidelines for the choice of geometric registra-
tion configuration.

Keywords Survey; Review; Iterative Closest Point algorithm; Point

set registration; Geometric registration; Mobile robotics; Laser odom-
etry; Search and Rescue; Inspection; Environmental monitoring; Au-
tonomous Driving.

F. Pomerleau, F. Colas and R. Siegwart. A Review of Point Cloud Registration

Algorithms for Mobile Robotics. Foundations and Trends R in Robotics, vol. 4,
no. 1, pp. 1–104, 2013.
DOI: 10.1561/2300000035.
 Full text available at: http://dx.doi.org/10.1561/2300000035

1
Twenty years of Iterative Closest Point (ICP):
The Legacy

The scope of this work is to present registration algorithms and their

use in mobile robotics. Registration algorithms associate sets of data
into a common coordinate system by minimizing the alignment error.
This allows to integrate data from different sources into a bigger model.
Although they can be quite an abstract and technical concept,
registration solutions already had an impact on the artistic field and
popular culture. Photographers proficiently use image registration to
build photograph composites achieving different looks-and-feels. The
Brenizer method is an exemplary technique that is applied to achieve
dramatic depth of field using panoramic image stitching (Figure 1.1 -
Top). Another example is High Dynamic Range (HDR) photographs,
where multiple images at different exposure levels need to be precisely
overlaid to retrieve details in shaded and highlighted areas (Figure 1.1
- Bottom). Nowadays, even the latest cellphones have the capacity to
build panoramic images from a series of pictures taken based on a vi-
sual guidelines that direct the user to move the camera viewfinder at
the optimal position for the next picture. As for the specific case of
3D mapping application, cinematographers are depicting possible uses
of registration algorithms in several recent science fiction movies. For

2
 Full text available at: http://dx.doi.org/10.1561/2300000035

instance, in the remake of Total Recall (Colombia Pictures, 2012), an

armed intervention team employed an array of hundreds of tiny cam-
eras in a dangerous room leading to a 3D reconstruction of the area
used to monitor potential threats within couple of seconds. Another
closely related potential application was the used by a geologist of fly-
ing drones carrying laser rangefinders to explore an alien facility in
Prometheus (Twentieth Century Fox, 2012).

Figure 1.1: Example of image registrations used in photography. Top: Brenizer

method using the open source software Hugin to stitch multiple images. Bottom:
HDR composite of the San Francisco harbor using the open source software Lumi-
nance HDR to overlay three images.
 Full text available at: http://dx.doi.org/10.1561/2300000035

4 Twenty years of ICP: The Legacy

More at the research level, current applications include: robotic

exploration in harsh environments, organ reconstruction to improve
medical diagnostics and object reconstruction for reverse engineering.
Although registration using 2D images can be part of the same group
of solutions, we focus on systems where depth information is already
available (e.g., from laser rangefinders) and is mainly used for resolving
misalignment error. We refer to the latter type as geometric registra-
tion 1 . However, some parallels with image registrations will be made
throughout this work when relevant.
A simplified example of geometric registration is illustrated in Fig-
ure 1.2. A scene with a large tree, a lamppost and a bench was scanned
using a laser rangefinder from two different poses. As laser points are
indistinguishable, only their location information is available to resolve
the alignment error. In that example, the point cloud in light green and
with the horizontal ground is used as our fixed reference coordinates.
Figure 1.2-Left shows the starting position of the two scans. The over-
laid point cloud in dark blue has a misalignment error shifting it to
the left with a tilt angle. This initial misalignment is represented with
dark red lines in Figure 1.2-Middle. Although all individual points are
similar, their proximity to other points gives enough information to
automatically align the two point clouds (Figure 1.2-Right).

Figure 1.2: Examples of geometric registration between a reference point cloud

(light green points) and a reading point cloud (dark blue points). Left: Initial position
of the two point clouds. Middle: Alignment error (dark red lines). Right: Final
alignment of the two point clouds.

1
In general, image registration often has access to labelled points, which is less
the case for geometric registration, either in 2D or 3D.
 Full text available at: http://dx.doi.org/10.1561/2300000035

1.1. Early Solutions 5

1.1 Early Solutions

As an interesting historical note, in an early publication by Hurley and

Cattell [1962], registration is presented as an Orthogonal Procrustes
problem. The name Procrustes refering to a bandit from the Greek
mythology who made his victims fit on his bed by either stretching
their limbs or cutting them off. Theseus eventually defeated Procrustes
using the same violent procedure (Figure 1.3). Nowadays, the reference
to the Orthogonal Procrustes problem is not often used in the scientific
literature, but it illustrates well the idea.

Figure 1.3: Theseus adjusting Procrustes to the size of his bed. Photograph pro-
vided by Marie-Lan Nguyen / Wikimedia Commons.

While working more specifically on 3D-shape primitives, Faugeras

and Hebert [1986] defined closed-form distances to minimize point-to-
point and plane-to-plane alignment error. The proposed method solved
translation and rotation as a two-step procedure. Later, a solution pro-
posed by Walker et al. [1991] resolved together rotation and translation
error using dual quaternions. The registration problem concretizes it-
self further in a survey of geometric matching algorithms and geometric
representations for point sets, curves, surfaces, volumes, and their re-
spective space-time trajectories [Besl, 1988]. At this time, the main ex-
 Full text available at: http://dx.doi.org/10.1561/2300000035

6 Twenty years of ICP: The Legacy

perimental validation was using Computer-aided design (CAD) models

with simple shapes. The first mention of the name ICP2 was proposed
by Besl and McKay [1992]. They expressed the problem as follows:

“Given 3-D data in a sensor coordinate system, which de-

scribes a data shape that may correspond to a model shape,
and given a model shape in a model coordinate system in a
different geometric shape representation, estimate the op-
timal rotation and translation that aligns, or registers, the
model shape and the data shape minimizing the distance
between the shapes and thereby allowing determination of
the equivalence of the shapes via a mean-square distance
metric.”

In their work, the proof of the solution convergence is demonstrated

under the assumption that the number of associated points, or their
weight, remains constant. Unless two identical shapes are registered
together, outliers that are not present in both shapes need to be iden-
tified. This problems is observed by Champleboux et al. [1992] while
developing early registration solutions for medical applications. They
report failures when wrong initial transformations are used in com-
bination with scans having low overlap ratio. During the same years,
Chen and Medioni [1991] work with dense laser scans of statues and,
shortly later, scans of tooth mockups [Chen and Medioni, 1992]. They
propose a registration solution based on the minimization of point-to-
plane alignment errors, which is still quite often used nowadays.
Even though a large volume of theoretical works was published on
advanced geometric primitives (e.g., planes, curves, quadrics), Zhang
[1994] states that primitives derived from points are too sensitive to
noise and are not stable in moving systems with current (1994) sens-
ing capabilities. Thus, he concludes that points were more reliable.
Zhang [1994] pioneers the idea of using ICP-based solutions for out-
door robotic applications. He proposes a generic framework for sym-
metric match, but considers only one direction of registration as an
2
In the remainder of this review, ICP and geometric registration have the same
generic meaning.
 Full text available at: http://dx.doi.org/10.1561/2300000035

1.2. Division and Explosion of the Field 7

approximation to save computation costs. He highly emphasizes the

importance of removing spurious pairs and gives the first character-
ization of fast subsampling solutions. In addition, he highlighted the
fact that outlier rejection is required for robotic applications, and that
the proof of ICP convergence stated by Besl and McKay [1992] can-
not hold for most of the robotics applications. In the outlook section
of his work, he already mentions the use of uncertainty on the initial
alignment, based on Kalman filters and Mahalanobis distance, and the
need to handle dynamic elements.

1.2 Division and Explosion of the Field

Within only two years, four main application types already emerged
from the possibilities to register 3D point clouds: object reconstructions
[Chen and Medioni, 1991], non-contact inspections [Besl and McKay,
1992], medical and surgery support [Champleboux et al., 1992] and au-
tonomous vehicle navigation [Zhang, 1993]. Publications in specialized
journals for computer vision, robotics and medical imaging slowly di-
vided the types of interesting problems to be solved. We can still read
in current literature that the credits for being the first article to provide
a solution differ from authors in different fields.
The field of registration crystalized with its first survey on med-
ical image registration covering the years 1993 to 1998 [Maintz and
Viergever, 1998]. It took 12 years for a specialized survey of 3D regis-
tration in computer vision to appear [Bowyer et al., 2006]. This work
intends to be the first large scale review adapted for Robotics applica-
tion.
ICP is a popular algorithm due to its simplicity: its general idea is
easy to understand and to implement. However, the basic algorithm
only works well in ideal cases. This led to hundreds3 of variations
around the original algorithm that were published on numerous dif-
ferent experimental scenarios (see Figure 1.4). This highlights both the
usefulness of ICP and the difficulty to find a single versatile version.
3
Close to 450 papers based on IEEE Xplore and around 1350 based on Scopus,
between 1992 and 2013.
 Full text available at: http://dx.doi.org/10.1561/2300000035

8 Twenty years of ICP: The Legacy

180

144
Number of papers

108

72

36

0
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
Years

Figure 1.4: Evolution of the number of publications over the years. Results were
obtained for the keywords Iterative Closest Point appearing in the abstract or the
title of publications. The dark blue area is computed based on IEEE Xplore database
and the light green area from the Scopus database.

IEEE Xplore Advance search request: ( "Document Title":"Iterative Closest Point" OR "Abstract":"Iterativ

In Figure 1.4, one can observe an increasing number of publications

Scopus Your query: TITLE-ABS-KEY("iterative closest point")
appearing around the year 2000. In robotics, this coincides with the ad-
vent of a 2D solution for pose estimations demonstrated with a SICK
rangefinder [Lu and Milios, 1997] and of the basis of Simultaneous Lo-
calization and Mapping (SLAM) algorithms [Thrun et al., 1998]. Prior
to the arrival of the SICK LMS-200 in robotics [Pfister et al., 2002],
most of the sensors used were custom-made. This situation renders ex-
periments difficult to replicate by other researchers. In those years, 2D
lasers appeared as a viable solution for navigation over sonars, which
were traditionally used [Thrun et al., 1998]. The 3D real-time applica-
tions were then not accessible due to high computation costs leading
to an increased research focus toward 2D solutions for autonomous
navigation, while other fields continued in 3D. At the same time in
computer vision, the seminal work of Rusinkiewicz and Levoy [2001]
on ICP algorithm comparisons led to significant progress in the scan
 Full text available at: http://dx.doi.org/10.1561/2300000035

1.2. Division and Explosion of the Field 9

registration field. The experiments employed simulated 3D scans, high-

lighting different spatial constraints and sensor noises. Results mainly
focused on the rapidity of convergence and the final precision of dif-
ferent solutions helping to select more appropriate solutions in further
applications.
With the arrival of more standard sensors, researchers in robotics
pushed the 2D registration algorithms so they could deal with larger
environments with faster platforms [Bosse et al., 2004] and 3D slowly
came back [Nüchter et al., 2004]. Since no comparison framework ex-
ists, the selection of an appropriate variant for particular experimental
conditions is difficult. This is a major challenge because registration is
at the front-end of the mapping pipeline, and the arbitrary nature of
its selection affects the results of all subsequent steps of more advanced
robotic tasks. Even the early work of Eggert et al. [1998] highlights the
difficulty to compare with other solutions given the lack of metric over
common data sets. In their survey, Maintz and Viergever [1998] point
the fact that proper accuracy studies are just starting; the problem be-
ing that the results provided are too specific. In addition, they highlight
the imprecise use of the terms accuracy, precision and robustness. They
suggest to set up public databases and validation protocols, but foresee
logistic, costs and efforts as incoming problem to those solutions.
Recently, the demand for a stronger experimental methodology in
robotics was also stressed by Amigoni et al. [2009]. The authors sur-
vey different SLAM publications to highlight proper evaluation met-
rics that are applied to SLAM algorithms. Three principles of an
experimental methodology in science (i.e, comparison, reproducibil-
ity/repeatability and justification/explanation) are translated in re-
quirements for stronger SLAM results. As stated in their paper, a
sound methodology should allow researchers to gain an insight about
intrinsic (e.g., computational time, parameters used, parameter behav-
iors) and extrinsic (e.g., accuracy, precision) quantities. The authors
report that, even though comparisons between algorithms are present
in SLAM publications, very few researchers can reuse the same proto-
col and directly compare their results without having to re-implement
other solutions.
 Full text available at: http://dx.doi.org/10.1561/2300000035

10 Twenty years of ICP: The Legacy

With the introduction of the Microsoft Kinect in 2010, another wave

of publications is expected, similar to what was observed following the
widespread utilization of SICK rangefinders. The Kinect is a handheld
camera sensor connected via USB to a computer that produces both
depth and color readings. Such RGB-D sensors augment accessibility
to object modeling and indoor mapping research [Henry et al., 2012].
This also opens the door to a mix of hybrid algorithms using features
and descriptors without the need of expertise in sensor calibration.
RGB-D cameras have different characteristics than laser-based sensors,
such as a higher density of points at a higher frequency but covering
a more restricted Field of View (FoV). A smaller FoV means less
time to compute the registration before the sensor trajectory reduce
the overlap to an unusable range. Having access to a higher frame
rate with an optimized ICP solution shows that hand-waved sensor
trajectory was trackable with real-time constraints [Pomerleau et al.,
2011]. The Velodyne HDL-64E, first commercialized for the DARPA
Urban Challenge in 2007, optimized its FoV to cover the expected
trajectory of a ground vehicle. To cope with the high speed of a car, the
sensor delivered a high data rate at 1.3 M points per second, bring the
real-time constraint to another level. Those two sensors were the latest
publication catalysts for the field of registration in mobile robotics,
field often modulated by the development of new hardwares.

1.3 Algorithm Overview

The aim of geometric registration is to be able to represent a shape,

called reading, in the same coordinate frame as another, called
reference. This is equivalent to finding the transformation of reading
that best aligns it to reference.
A shape S is a set of points including both geometric and non-
geometric information. Geometric information is affected by a spatial
transformation; this part of the dimension of a point will be called a
feature. Features are typically coordinates of points, surface normals or
tangent vectors. Non-geometric information is not affected by spatial
 Full text available at: http://dx.doi.org/10.1561/2300000035

1.3. Algorithm Overview 11

transformation; this part of the dimension of a point will be called a

descriptor. Descriptors can be color, temperature, identifiers, etc.
Most algorithms actually apply some filters on the shapes in order
to help the registration. There are mainly two uses of such filters. The
first one is to remove some points that do not bring any valuable infor-
mation for the registration. As the complexity of the algorithm is linear
in the number of points, reducing this number can have a significant
impact on the time of registration. The second use of filters can be to
add information to the point. The typical example is the inference of
local structural properties of the shape, such as normal information or
curvature. This information, which is usually not present in the raw
sensor data, can allow for better registration through a more precise
association of the points, or the computation of the error to minimize.
More formally, let AP be the shape representing reading in a coor-
dinate frame A and BQ the shape representing reference in its coordi-
nate frame B. The aim of registration is to estimate the transformation
0
A T by minimizing an error function error(P , Q):
B

    
B A
A T̂ = arg min error T P , BQ (1.1)
T

where T (S) is the application of the geometric transformation T to the

shape S.
One specificity of geometric registration is that the error function is
computed on pairs of points that have been associated between the two
shapes. The classical association is done by finding the closest point in
reference of each point in reading. Ideally the association should be
between points that, when the two shapes are aligned, are the closest
in position. This problem is called data association, point matching,
correspondence finding depending on the literature. Association solving
can be done purely on the features but can also be improved by using
the descriptors.
Formally, let M = match(P, Q) = {(p, q) : p ∈ P, q ∈ Q} be the
set of matches between P and Q. The error function is then of the
form:
X
error(P, Q) = d(p, q).
(p,q)∈M
 Full text available at: http://dx.doi.org/10.1561/2300000035

12 Twenty years of ICP: The Legacy

In order to make this error function more robust, outliers are some-
times identified and removed from the list of matches. In addition,
weights W = outlier(M) = {w(p, q) : ∀(p, q) ∈ M} can be associated
to the matches so as to increase or decrease their influence in the error
function: X
error(P, Q) = w(p, q) d(p, q).
(p,q)∈M

It is clear that minimizing this error function with an ideal associ-

ation yields the best estimate for ABT (Equation 1.1). However, unless
the descriptors are discriminative enough (as with visual descriptors),
the association can generally not be perfectly solved. The idea of ICP
is that even with an imperfect association, minimizing the error yields
a better estimates that, in turn, allows for better association. Con-
cretely, the idea is to build a sequence of transformations i−1iT that
are successively applied to P. At a given iteration, a set of matches
Mi is computed from the given relative position of the points. Then,
based of those matches, a new transformation i+1 i T is computed by
minimizing the error:
    
i
i+1
iT ← arg min error T P 0 , Q0 . (1.2)
T

Finally, the estimate of the transformation between the two original

shapes is the composition of all intermediary transformations:
 
B i
A T̂ = i−1 T ◦ Tinit (1.3)
i

i 3 2
where i−1 T = ··· ◦ 2T ◦ 1T is the iterative composition of the
i
transformations, and Tinit an initial transformation.
The generic procedure is summarized in Algorithm 1 and shown as
a chart in Figure 1.5.

1.4 Overview of the Review

ICP is a framework where multiple variations and algorithms can be

used to resolve geometric registration problems. In the light of this
large corpus of work related to ICP and more generally to geometric
 Full text available at: http://dx.doi.org/10.1561/2300000035

1.4. Overview of the Review 13

Algorithm 1 Summary of ICP algorithm.

Require: AP . reading
Require: BQ . reference
Require: Tinit . initial transformation
A 0
P ← datafilter( P )
A . data filters
B 0
Q ← datafilter( BQ ) . data filters
iT ← T
i−1 init
repeat
i 0
P ← i−1iT ( i−1P 0 ) . move reading
Mi ← match( iP 0 , Q0 ) . associate points
Wi ← outlier(Mi)     . filter outliers
i+1 i 0 0
i T ← arg min error T P ,Q
T
until convergence
 
i
A T̂ i−1 T ◦ Tinit
Ensure: B =
i

Initial
Transformation

Data Outlier
Reading
Filter(s) Filter(s)
Association Error
Solver Minimization

Data Outlier
Reference
Filter(s) Filter(s)

Legend:

Features Descriptors Assumptions Assumptions

Figure 1.5: Generic scheme proposed for registration algorithms.

 Full text available at: http://dx.doi.org/10.1561/2300000035

14 Twenty years of ICP: The Legacy

registration, we present a general framework to classify the existing so-

lutions. We believe that after 20 years of new registration algorithms, it
is time to evaluate what works best for which robotic systems. There-
fore, our contributions aim at strengthening the current methodology
and bring deeper analysis of current solutions. The timing is appro-
priate for such study given that computational power is now sufficient
to support registration on embedded systems in real-time [Pomerleau
et al., 2011]. Also, new advancements in electronics have improved the
accuracy and speed of sensors. Improvements in battery technology
have enabled longer autonomous operation time. Most importantly, re-
searchers face a plethora of solutions from which a definition of usable
solutions can be out of reach. This situation impedes the robotic field
to progress on algorithms that rely on registration (e.g., path planning,
autonomous exploration).
This review addresses this problem and is structured in two main
sections:

Section 2 presents a literature review of different solutions with the

aim to express ICP solutions in a common framework and vali-
date our generic scheme proposed in Figure 1.5.

Section 3 describes case studies using five different robotic platforms.

The requirements of each application are explained with some in-
sight on how to tune parameters for specific applications. Those
applications cover Search & Rescue activities, industrial inspec-
tion, shore monitoring and autonomous driving.

All sections close with a discussion in addition to a short summary.

The main observations of those sections are recapitulated in Section 4
along with final remarks.
 Full text available at: http://dx.doi.org/10.1561/2300000035

References

Francesco Amigoni, Monica Reggiani, and Viola Schiaffonati. An insightful

comparison between experiments in mobile robotics and in science. Au-
tonomous Robots, 27(4):313–325, August 2009.
Leopoldo Armesto, Javier Minguez, and Luis Montesano. A generalization of
the metric-based Iterative Closest Point technique for 3D scan matching.
In Robotics and Automation, 2010. Proceedings of the IEEE International
Conference on, pages 1367–1372, 2010.
K S Arun, T S Huang, and S D Blostein. Least-Squares Fitting of Two 3-D
Point Sets. Pattern Analysis and Machine Intelligence, IEEE Transactions
on, 9(5):698–700, 1987.
Sunil Arya and David Mount. Approximate nearest neighbor queries in fixed
dimensions. In Discrete Algorithms, 1993. Proceedings of the 4th Annual
ACM-SIAM Symposium on, pages 271–280. Department of Computer Sci-
ence, University of Maryland, College Park, Maryland, 20742 Department
of Computer Science and Institute for Advanced Computer Studies, Uni-
versity of Maryland, College Park, Mary- land, 20742, 1993.
R Benjemaa and F Schmitt. Fast global registration of 3D sampled surfaces
using a multi-z-buffer technique. In 3-D Digital Imaging and Modeling,
1997. Proceedings of the International Conference on Recent Advances in,
pages 113–120, 1997.
R Bergevin, M Soucy, H Gagnon, and D Laurendeau. Towards a general multi-
view registration technique. Pattern Analysis and Machine Intelligence,
IEEE Transactions on, 18(5):540–547, 1996.

95
 Full text available at: http://dx.doi.org/10.1561/2300000035

96 References

P Besl and H McKay. A method for registration of 3-D shapes. Pattern

Analysis and Machine Intelligence, IEEE Transactions on, 14(2):239–256,
February 1992.
P J Besl. Geometric modeling and computer vision. In Proceedings of the
IEEE, pages 936–958, 1988.
Michael Bosse and R Zlot. Map Matching and Data Association for Large-
Scale Two-dimensional Laser Scan-based SLAM. The International Journal
of Robotics Research, 27(6):667–691, June 2008.
Michael Bosse and R Zlot. Keypoint design and evaluation for place recog-
nition in 2D lidar maps. Robotics and Autonomous Systems, 57(12):1211–
1224, 2009a.
Michael Bosse and Robert Zlot. Continuous 3D scan-matching with a spinning
2D laser. In Robotics and Automation, 2009. Proceedings of the IEEE
International Conference on, pages 4312–4319, 2009b.
Michael Bosse, Paul Newman, John Leonard, and Seth Teller. Simultaneous
Localization and Map Building in Large-Scale Cyclic Environments Using
the Atlas Framework. The International Journal of Robotics Research, 23
(12):1113–1139, December 2004.
Kevin W Bowyer, Kyong Chang, and Patrick Flynn. A survey of approaches
and challenges in 3D and multi-modal 3D+2D face recognition. Computer
Vision And Image Understanding, 101(1):1–15, January 2006.
A Censi. Scan matching in a probabilistic framework. In Robotics and Automa-
tion, 2006. Proceedings of the IEEE International Conference on, pages
2291–2296, 2006.
A Censi. An ICP variant using a point-to-line metric. In Robotics and Automa-
tion, 2008. Proceedings of the IEEE International Conference on, pages
19–25, 2008.
G Champleboux, S Lavallee, R Szeliski, and L Brunie. From accurate range
imaging sensor calibration to accurate model-based 3D object localization.
In Computer Vision and Pattern Recognition, 1992. Proceedings of the 1992
IEEE Computer Society Conference on, pages 83–89, 1992.
Y Chen and G Medioni. Object modeling by registration of multiple range
images. In Robotics and Automation, 1991. Proceedings of the IEEE Inter-
national Conference on, pages 2724–2729. IEEE Comput. Soc. Press, April
1991.
Yang Chen and Gérard Medioni. Object modelling by registration of multiple
range images. Image And Vision Computing, 10(3):145–155, April 1992.
 Full text available at: http://dx.doi.org/10.1561/2300000035

References 97

D Chetverikov, D Svirko, D Stepanov, and P Krsek. The Trimmed Iterative

Closest Point algorithm. In Pattern Recognition, 2002. Proceedings of the
16th International Conference on, pages 545–548, 2002.
Francis Colas, Srivatsa Mahesh, François Pomerleau, Ming Liu, and Roland
Siegwart. 3D path planning and execution for search and rescue ground
robots. In Intelligent Robots and Systems (IROS), 2013 IEEE/RSJ Inter-
national Conference on. IEEE, 2013.
J Diebel, K Reutersward, S Thrun, J Davis, and R Gupta. Simultaneous
localization and mapping with active stereo vision. In Intelligent Robots
and Systems, 2004. Proceedings of the IEEE/RSJ International Conference
on, pages 3436–3443 vol.4, 2004.
S Druon, M J Aldon, and A Crosnier. Color constrained ICP for registration
of large unstructured 3d color data sets. In Information Acquisition, 2006.
Proceedings of the IEEE International Conference on, pages 249–255. IEEE,
2006.
David W Eggert, Adele Lorusso, and Robert B Fisher. Estimating 3-D rigid
body transformations: A comparison of four major algorithms. Machine
Vision and Applications, 9(5-6):272–290, 1997.
David W Eggert, Andrew W Fitzgibbon, and Robert B Fisher. Simultaneous
Registration of Multiple Range Views for Use in Reverse Engineering of
CAD Models. Computer Vision And Image Understanding, 69(3):253–272,
February 1998.
N Fairfield and D Wettergreen. Evidence grid-based methods for 3D map
matching. In Robotics and Automation, 2009. Proceedings of the IEEE
International Conference on, pages 1637–1642, May 2009.
O D Faugeras and M Hebert. The Representation, Recognition, and Locating
of 3-D Objects. The International Journal of Robotics Research, 5(3):27–52,
September 1986.
J Feldmar and N Ayache. Locally affine registration of free-form surfaces. In
Computer Vision and Pattern Recognition, 1994. Proceedings of the IEEE
Computer Society Conference on, pages 496–501, 1994.
Jacques Feldmar and Nicholas Ayache. Rigid, affine and locally affine regis-
tration of free-form surfaces. International Journal of Computer Vision, 18
(2):99–119, May 1996.
O Fluck, C Vetter, W Wein, A Kamen, B Preim, and R Westermann. A survey
of medical image registration on graphics hardware. Computer Methods and
Programs in Biomedicine, 104(3):e45–e57, December 2011.
 Full text available at: http://dx.doi.org/10.1561/2300000035

98 References

H Gagnon, M Soucy, R Bergevin, and D Laurendeau. Registration of mul-

tiple range views for automatic 3-D model building. In Computer Vision
and Pattern Recognition, 1994. Proceedings of the IEEE Computer Society
Conference on, pages 581–586, 1994.
N Gelfand, L Ikemoto, S Rusinkiewicz, and M Levoy. Geometrically stable
sampling for the ICP algorithm. In 3-D Digital Imaging and Modeling,
2003. Proceedings of the Fourth International Conference on, pages 260–
267, 2003.
G Godin, M Rioux, and R Baribeau. Three-dimensional registration using
range and intensity information. In Videometric III. Proceedings of SPIE
Conference on, pages 279–290, 1994.
Steven Gold, Anand Rangarajan, Chien-Ping Lu, Suguna Pappu, and Eric
Mjolsness. New algorithms for 2D and 3D point matching: pose estimation
and correspondence. Pattern Recognition, 31(8):1019–1031, August 1998.
Giorgio Grisetti, C Stachniss, and W Burgard. Improving Grid-based SLAM
with Rao-Blackwellized Particle Filters by Adaptive Proposals and Selective
Resampling. In Robotics and Automation, 2005. Proceedings of the IEEE
International Conference on, pages 2432–2437, 2005.
Giorgio Grisetti, S Grzonka, C Stachniss, P Pfaff, and W Burgard. Efficient es-
timation of accurate maximum likelihood maps in 3D. In Intelligent Robots
and Systems, 2007. Proceedings of the IEEE/RSJ International Conference
on, pages 3472–3478, 2007.
P Henry, M Krainin, E Herbst, X Ren, and D Fox. RGB-D mapping: Using
Kinect-style depth cameras for dense 3D modeling of indoor environments.
The International Journal of Robotics Research, 31(5):647–663, April 2012.
Gregory Hitz, François Pomerleau, Marie-Eve Garneau, Cedric Pradalier,
Thomas Posch, Jakob Pernthaler, and Ronald Siegwart. Autonomous
Inland Water Monitoring: Design and Application of a Surface Vessel.
Robotics & Automation Magazine, IEEE, 19(1):62–72, March 2012.
Gregory Hitz, François Pomerleau, Francis Colas, and Roland Siegwart. State
Estimation for Shore Monitoring Using an Autonomous Surface Vessel. In
International Symposium on Experimental Robotics (ISER), pages 1–15,
Marrakech/Essaouira, June 2014.
Berthold K P Horn. Closed-form solution of absolute orientation using unit
quaternions. Journal of the Optical Society of America A, 4(4):629, 1987.
Berthold K P Horn, Hugh M Hilden, and Shahriar Negahdaripour. Closed-
form solution of absolute orientation using orthonormal matrices. Journal
of the Optical Society of America A, 5(7):1127–1135, 1988.
 Full text available at: http://dx.doi.org/10.1561/2300000035

References 99

John R Hurley and Raymond B Cattell. The Procrustes program: Producing

direct rotation to test a hypothesized factor structure. Behavioral Science,
7(2):258–262, April 1962.
A E Johnson and Sing Bing Kang. Registration and integration of textured
3-D data. In 3-D Digital Imaging and Modeling, 1997. Proceedings of the
International Conference on Recent Advances in, pages 234–241, 1997.
T Jost and H Hügli. Fast ICP algorithms for shape registration. In Pattern
Recognition, 2002. Proceedings of the 24th DAGM German Symposium on,
pages 91–99. Pattern Recognition Group, Institute of Microtechnology, Uni-
versity of Neuchâtel, Breguet 2, CH-2000 Neuchâtel, Switzerland, 2002.
T Jost and H Hugli. A multi-resolution scheme ICP algorithm for fast shape
registration. In 3D Data Processing Visualization and Transmission, 2002.
Proceedings of the First International Symposium on, pages 540–543, 2002.
T Jost and H Hugli. A multi-resolution ICP with heuristic closest point search
for fast and robust 3D registration of range images. In 3-D Digital Imaging
and Modeling, 2003. Proceedings of the Fourth International Conference
on, pages 427–433, 2003.
D Kim. A Fast ICP Algorithm for 3-D Human Body Motion Tracking. Signal
Processing Letters, IEEE, 17(4):402–405, 2010.
G J M Kruijff, M Janicek, S Keshavdas, B Larochelle, H Zender, N J J M
Smets, T Mioch, M A Neerincx, J van Diggelen, Francis Colas, M Liu,
F Pomerleau, Roland Siegwart, V Hlavac, T Svoboda, T Petricek, M Rein-
stein, K Zimmerman, F Pirri, M Gianni, P Papadakis, A Sinha, P Balmer,
N Tomatis, R Worst, T Linder, H Surmann, V Tretyakov, H Surmann,
S Corrao, S Pratzler-Wanczura, and M Sulk. Experience in System De-
sign for Human-Robot Teaming in Urban Search and Rescue. In Field and
Service Robotics, pages 1–14, Matsushima, Japan, July 2012.
Vladimír Kubelka, Lorenz Oswald, François Pomerleau, Francis Colas, Tomáš
Svoboda, and Michal Reinstein. Robust data fusion of multi-modal sensory
information for mobile robots. Journal of Field Robotics, in press, 2014.
P Kumari, R Shrestha, and B Carter. Registration of LiDAR data through
stable surface matching. In Geoinformatics, 2009. Proceedings of the 17th
International Conference on, pages 1–5, 2009.
P Lamon, S Kolski, and Roland Siegwart. The SmartTer-a vehicle for fully au-
tonomous navigation and mapping in outdoor environments. In Proceedings
of CLAWAR, 2006.
 Full text available at: http://dx.doi.org/10.1561/2300000035

100 References

Ce Li, Jianru Xue, Shaoyi Y Du, and Nanning Zheng. A Fast Multi-Resolution
Iterative Closest Point Algorithm. In Pattern Recognition, 2010. Proceed-
ings of the Chinese Conference on, pages 1–5, 2010.
Yonghuai Liu. Automatic Range Image Registration in the Markov Chain.
Pattern Analysis and Machine Intelligence, IEEE Transactions on, 32(1):
12–29, 2010.
D Lowe. Distinctive image features from scale-invariant keypoints. Interna-
tional Journal of Computer Vision, 60(2):91–110, 2004.
F Lu and E Milios. Globally Consistent Range Scan Alignment for Environ-
ment Mapping. Autonomous Robots, 4(4):333–349, 1997.
M Magnusson, Andreas Nüchter, C Lorken, A Lilienthal, and J Hertzberg.
Evaluation of 3D registration reliability and speed - A comparison of ICP
and NDT. In Robotics and Automation, 2009. Proceedings of the IEEE
International Conference on, pages 3907–3912, 2009.
J B Antoine Maintz and Max A Viergever. A survey of medical image regis-
tration. Medical Image Analysis, 2(1):1–36, March 1998.
P Markelj, D Tomaževič, B Likar, and F Pernuš. A review of 3D/2D regis-
tration methods for image-guided interventions. Medical Image Analysis,
16(3):642–661, April 2012.
T Masuda. Generation of geometric model by registration and integration
of multiple range images. In 3-D Digital Imaging and Modeling, 2001.
Proceedings of the Third International Conference on, pages 254–261, 2001.
T Masuda, K Sakaue, and N Yokoya. Registration and integration of multiple
range images for 3-D model construction. In Pattern Recognition, 1996.
Proceedings of the 13th International Conference on, pages 879–883 vol.1,
1996.
C R Jr Maurer, G B Aboutanos, B M Dawant, R J Maciunas, and J M
Fitzpatrick. Registration of 3-D images using weighted geometrical features.
Medical Imaging, IEEE Transactions on, 15(6):836–849, 1996.
G Medioni, C K Tang, and M S Lee. Tensor voting: Theory and applications.
In Reconnaissance des formes et Intelligence Artificielle, 2000. Proceedings
of the Conference on, 2000.
Helena Mitasova, Margery F Overton, Juan José Recalde, David J Bernstein,
and Christopher W Freeman. Raster-Based Analysis of Coastal Terrain
Dynamics from Multitemporal Lidar Data. Journal of Coastal Research,
252:507–514, March 2009.
 Full text available at: http://dx.doi.org/10.1561/2300000035

References 101

E Mortensen, Hongli Deng, and L Shapiro. A SIFT descriptor with global

context. In Computer Vision and Pattern Recognition, 2005. Proceedings
of the IEEE Computer Society Conference on, pages 184–190 vol. 1, 2005.
Andreas Nüchter, H Surmann, K Lingemann, J Hertzberg, and S Thrun. 6D
SLAM with an application in autonomous mine mapping. In Robotics and
Automation, 2004. Proceedings of the IEEE International Conference on,
pages 1998–2003 Vol.2, 2004.
Andreas Nüchter, K Lingemann, J Hertzberg, and H Surmann. 6D SLAM with
approximate data association. In Advanced Robotics, 2005. Proceedings of
the 12th International Conference on, pages 242–249, 2005.
Andreas Nüchter, K Lingemann, and J Hertzberg. Cached k-d tree search for
ICP algorithms. In 3-D Digital Imaging and Modeling, 2007. Proceeding of
the Sixth International Conference on, pages 419–426, 2007.
Ye Pan, Bo Dai, and Qicong Peng. Fast and robust 3D face matching ap-
proach. In Image Analysis and Signal Processing, 2010. Proceedings of the
Second International Conference on, pages 195–198, 2010.
S T Pfister, K L Kriechbaum, S I Roumeliotis, and J W Burdick. Weighted
range sensor matching algorithms for mobile robot displacement estimation.
In Robotics and Automation, 2002. Proceedings of the IEEE International
Conference on, 2002.
J P W Pluim, J B A Maintz, and M A Viergever. Mutual-information-based
registration of medical images: a survey. Medical Imaging, IEEE Transac-
tions on, 22(8):986–1004, 2003.
François Pomerleau, Francis Colas, François Ferland, and F Michaud. Rela-
tive motion threshold for rejection in ICP registration. Field and Service
Robotics, pages 229–238, 2010.
François Pomerleau, Stéphane Magnenat, Francis Colas, Ming Liu, and
Roland Siegwart. Tracking a depth camera: Parameter exploration for fast
ICP. In Intelligent Robots and Systems, 2011. Proceedings of the IEEE/RSJ
International Conference on, pages 3824–3829, 2011.
François Pomerleau, Andreas Breitenmoser, Ming Liu, Francis Colas, and
Roland Siegwart. Noise Characterization of Depth Sensors for Surface In-
spections. In Applied Robotics for the Power Industry, 2012. Proceedings
of the Second International Conference on, pages 1–6, Zurich, Switzerland,
August 2012a.
François Pomerleau, M. Liu, Francis Colas, and Roland Siegwart. Challeng-
ing Data Sets for Point Cloud Registration Algorithms. The International
Journal of Robotics Research, September 2012b.
 Full text available at: http://dx.doi.org/10.1561/2300000035

102 References

François Pomerleau, Francis Colas, Roland Siegwart, and Stéphane Magnenat.

Comparing ICP Variants on Real-World Data Sets. Autonomous Robots,
34(3):133–148, February 2013.
François Pomerleau, Philipp Krüsi, Paul Furgale, and Roland Siegwart. Long-
term 3D map maintenance in dynamic environments. In 2014 IEEE Inter-
national Conference on Robotics and Automation (ICRA 2014), pages 1–8,
February 2014.
K Pulli. Multiview registration for large data sets. In 3-D Digital Imaging
and Modeling, 1999. Proceedings of the Second International Conference
on, pages 160–168, October 1999.
L Reyes, G Medioni, and E Bayro. Registration of 3D points using geometric
algebra and tensor voting. International Journal of Computer Vision, 2007.
S Rusinkiewicz and M Levoy. Efficient variants of the ICP algorithm. In 3-D
Digital Imaging and Modeling, 2001. Proceedings of the Third International
Conference on, pages 145–152, 2001.
J Salvi, C Matabosch, D Fofi, and J Forest. A review of recent range image
registration methods with accuracy evaluation. Image And Vision Com-
puting, 2007.
C Schutz, T Jost, and H Hugli. Multi-feature matching algorithm for free-
form 3D surface registration. In Pattern Recognition, 1998. Proceedings of
the 14th International Conference on, pages 982–984 vol.2, 1998.
Aleksandr Segal, Dirk Haehnel, and Sebastian Thrun. Generalized-ICP.
In Robotics: Science and Systems V. Proceedings of the Conference on,
page 21. Stanford University, 2009.
L Silva, O Bellon, and K Boyer. Precision range image registration using a
robust surface interpenetration measure and enhanced genetic algorithms.
Pattern Analysis and Machine Intelligence, IEEE Transactions on, 27(5):
762–776, 2005.
C Stewart, Chia-Ling Tsai, and B Roysam. The dual-bootstrap iterative clos-
est point algorithm with application to retinal image registration. Medical
Imaging, IEEE Transactions on, 22(11):1379–1394, 2003.
E Stumm, A Breitenmoser, François Pomerleau, C Pradalier, and Roland Sieg-
wart. Tensor-voting-based navigation for robotic inspection of 3D surfaces
using lidar point clouds. The International Journal of Robotics Research,
31(12):1465–1488, November 2012.
 Full text available at: http://dx.doi.org/10.1561/2300000035

References 103

Fabien Tâche, Wolfgang Fischer, Gilles Caprari, Roland Siegwart, Roland

Moser, and Francesco Mondada. Magnebike: A magnetic wheeled robot
with high mobility for inspecting complex-shaped structures. Journal of
Field Robotics, 26(5), May 2009.
Fabien Tâche, François Pomerleau, W Fischer, Gilles Caprari, F Mondada,
R Moser, and Roland Siegwart. MagneBike: Compact magnetic wheeled
robot for power plant inspection. In Applied Robotics for the Power Indus-
try, 2010. Proceedings of the First International Conference on, pages 1–2,
2010.
Fabien Tâche, François Pomerleau, Gilles Caprari, Roland Siegwart, Michael
Bosse, and Roland Moser. Three-Dimensional Localization for the Mag-
neBike Inspection Robot. Journal of Field Robotics, 28(2):180–203, 2011.
Sebastian Thrun, Wolfram Burgard, and Dieter Fox. A Probabilistic Ap-
proach to Concurrent Mapping and Localization for Mobile Robots. Ma-
chine Learning, 31(1/3):29–53, 1998.
C Tsai, C Li, G Yang, and K Lin. The Edge-Driven Dual-Bootstrap Iter-
ative Closest Point Algorithm for Registration of Multimodal Fluorescein
Angiogram Sequence. Medical Imaging, IEEE Transactions on, 29(3):636–
649, 2010.
T Tuytelaars and K Mikolajczyk. A survey on local invariant features. Foun-
dations and Trends in Computer Graphics and Vision, 2008.
Michael W Walker, Lejun Shao, and Richard A Volz. Estimating 3-D location
parameters using dual number quaternions. CVGIP: Image Understanding,
54(3):358–367, November 1991.
O Wulf, Andreas Nüchter, J Hertzberg, and B Wagner. Benchmarking urban
six-degree-of-freedom simultaneous localization and mapping. Journal of
Field Robotics, 25(3):148–163, 2008.
Kai M Wurm, Armin Hornung, Maren Bennewitz, Cyrill Stachniss, and Wol-
fram Burgard. OctoMap: A Probabilistic, Flexible, and Compact 3D Map
Representation for Robotic Systems. In Workshop on Best Practice in 3D
Perception and Modeling for Mobile Manipulation, 2010 (ICRA), 2010.
H Yoshitaka, K Hirohiko, O Akihisa, and Y Shin’ichi. Mobile Robot Local-
ization and Mapping by Scan Matching using Laser Reflection Intensity of
the SOKUIKI Sensor. In Industrial Electronics, 2006. Proceedings of the
IEEE 32nd Annual Conference on, pages 3018–3023, 2006.
Zhengyou Zhang. Point matching for registration of free-form surfaces. Lecture
Notes in Computer Science, 719:460–467, May 1993.
 Full text available at: http://dx.doi.org/10.1561/2300000035

104 References

Zhengyou Zhang. Iterative point matching for registration of free-form curves

and surfaces. International Journal of Computer Vision, 13(2):119–152,
October 1994.
T Zinsser, J Schmidt, and H Niemann. A refined ICP algorithm for robust
3-D correspondence estimation. In Image Processing, 2003. Proceedings of
the IEEE International Conference on, pages II–695–8 vol.3, 2003.
Barbara Zitová and Jan Flusser. Image registration methods: a survey. Image
And Vision Computing, 21(11):977–1000, 2003.
R Zlot and Michael Bosse. Place recognition using keypoint similarities in 2D
lidar maps. Experimental Robotics, 54:363–372, 2009.