Estimating pose uncertainty for surface registration

doi:10.1016/S0262-8856(97)00059-0

Image and Vision Computing

Volume 16, Issue 2, 20 February 1998, Pages 111-120

https://doi.org/10.1016/S0262-8856(97)00059-0 Get rights and content

Abstract

Accurate registration of surfaces is a common task in computer vision. Several algorithms exist to refine an approximate value for the pose to an accurate value. They are all more or less variants of the iterative closest point (ICP) algorithm of Besl and McKay. Up to now the problem of determining the uncertainty in the pose estimate when registering surfaces has not been solved. In this paper we provide a solution applicable when registering a noisy measured surface to an accurately known surface model. We contend that it is necessary to use the normal-projection ICP of Chen and Medioni to obtain a meaningful uncertainty estimate. Knowledge of the uncertainty provides a quantitative signal for cases of near degenerate surface shape where accurate pose estimation may not be possible. We introduce a new parameter called the registration index to give a simple means of quantifying the pose errors one might expect when registering a particular shape.

References (21)

Y. Chen et al.
Object modelling by registration of multiple range images
Image and Vision Computing
(1992)
K. Kanatani
Analysis of 3D rotation fitting
IEEE Transactions on Pattern Analysis and Machine Intelligence
(1994)
P.J. Besl et al.
A method for registration of 3D shapes
IEEE Transactions on Pattern Analysis and Machine Intelligence
(1992)
X. Pennec et al.
Validation of 3D registration methods based on points and frames
X. Pennec et al.
Validation of 3D registration methods based on points and frames
INRIA Tech Report RR2470
(1995)
X. Pennec, J.-P. Thirion, A framework for uncertainty and validation of 3D registration methods based on points and...
H.I. Christensen et al.
R.M. Haralick
Overview: computer vision performance characterization
S. Yi et al.
Error propagation in machine vision
Machine Vision and Applications
(1994)
D. Brujic, M. Ristic, Monte Carlo simulation and analysis of free form surface registration, Journal of Engineering...

There are more references available in the full text version of this article.

Cited by (16)

Global registration of large collections of range images with an improved Optimization-on-a-Manifold approach
2014, Image and Vision Computing
Citation Excerpt :
These methods are usually employed to improve the alignment obtained after the application of a coarse alignment technique. Many approaches to this problem are related to the Iterative Closest Point (ICP) technique, first introduced in [13,14], and to its efficient variants introduced afterwards [15–18]. The ICP technique is an iterative approach for which, at every iteration, a set of correspondences is established between two views, and used to estimate the rototranslation matrix that brings these views closer to alignment.
Concurrently obtaining an accurate, robust and fast global registration of multiple 3D scans is still an open issue for modern 3D modeling pipelines, especially when high metric precision as well as easy usage of high-end devices (structured-light or laser scanners) are required. Various solutions have been proposed (either heuristic, iterative and/or closed form solutions) which present some compromise concerning the fulfillment of the above contrasting requirements. Our purpose here, compared to existing reference solutions, is to go a step further in this perspective by presenting a new technique able to provide improved alignment performance, even on large datasets (both in terms of number of views and/or point density) of range images. Relying on the ‘Optimization-on-a-Manifold’ (OOM) approach, originally proposed by Krishnan et al., we propose a set of methodological and computational upgrades that produce an operative impact on both accuracy, robustness and computational performance compared to the original solution. In particular, always basing on an unconstrained error minimization over the manifold of rotations, instead of relying on a static set of point correspondences, our algorithm updates the optimization iterations with a dynamically modified set of correspondences in a computationally effective way, leading to substantial improvements in terms of registration accuracy and convergence trend. Other proposed improvements are directed to a substantial reduction of the computational load without sacrificing the alignment performance. Stress tests with increasing view misalignment allowed us to appreciate the convergence robustness of the proposed solution. Eventually, we demonstrate that for very large datasets a further computational speedup can be reached by the adoption of a hybrid (local heuristic followed by global optimization) registration approach.
A method for automated registration of unorganised point clouds
2008, ISPRS Journal of Photogrammetry and Remote Sensing
Citation Excerpt :
Some research about the estimation of position uncertainty in range sensors has been conducted. For example, registration of three-dimensional data (Johnson and Kang, 1997; Dorai et al., 1997; Pennec and Thirion, 1997; Stoddart et al., 1996), uncertainty of sampled three-dimensional surfaces (Tasdizen and Whitaker, 2003; Pauly et al., 2004; Whaite and Ferrie, 1991), range error analysis for different kinds of laser scanners (Blais et al., 2000) and surface normal estimation in noisy datasets (Mitra et al., 2004). Except for Tasdizen and Whitaker (2003), Mitra et al. (2004), Guehring (2001), Williams et al. (1999), Blais et al. (2000) and Whaite and Ferrie (1997), the others regard the measurement of the position of a point using laser scanners as a stochastic process with a three-dimensional, isotropic, and zero-mean Gaussian probability density function.
A registration method for unorganised point cloud datasets, the Geometric Primitive ICP with the RANSAC (GP-ICPR), which uses geometric primitives, neighbourhood search and the positional uncertainty of laser scanners is proposed. The change of geometric curvature and approximate normal vector of the surface formed by a point and its neighbourhood are used to search for possible corresponding points.
The GP-ICPR was tested with terrestrial laser scanner datasets in terms of its precision and accuracy. It is shown that the GP-ICPR improved the precision of the estimated relative transformation parameters by as much as a factor of 5, which provides a window of opportunity to utilise this automated registration method in practical applications such as terrestrial surveying and deformation monitoring.
Recursive scan-matching SLAM
2007, Robotics and Autonomous Systems
Citation Excerpt :
The averaged RMS error for the vertices gives an estimate of point alignment accuracy. Similarly, in [18] another parameter is defined called registration index. The registration index gives an indication of how well two surfaces may be registered.
This paper presents Scan-SLAM, a new generalization of simultaneous localization and mapping (SLAM). SLAM implementations based on extended Kalman filter (EKF) data fusion have traditionally relied on simple geometric models for defining landmarks. This limits EKF-SLAM to environments suited to such models and tends to discard much potentially useful data. The approach presented in this paper is a marriage of EKF-SLAM and scan correlation. Landmarks are no longer defined by analytical models; instead they are defined by templates composed of raw sensed data. These templates can be augmented as more data become available so that the landmark definition improves with time. A new generic observation model is derived that is generated by scan correlation, and this permits stochastic location estimation for landmarks with arbitrary shape within the Kalman filter framework. The statistical advantages of an EKF representation are augmented with the general applicability of scan matching. Scan matching also serves to enhance data association reliability by providing a shape metric for landmark disambiguation. Experimental results in an outdoor environment are presented which validate the algorithm.
Analysis of skin movement with respect to flexional bone motion using MR images of a hand
2006, Journal of Biomechanics
This study describes the methodology for analyzing skin movement with respect to the predefined flexional motion of the hand. The 3D in vivo geometric data were acquired through magnetic resonance (MR) scanning of multiple hands in various postures. Custom software was developed to segment MR images and to generate polygonal isosurfaces. A new bone coordinate system (BCS) was defined to describe the skin movement measured with the skin-surface markers. The BCS was defined by the surface registration technique, which could minimize the distance among the 3D polygonal surfaces of different postures.
MR images of six subjects in the 3 postures were captured with skin-surface markers for the experiment. Skin movement on the second metacarpal bone, which was fixed during the flexional motion, ranged from 1.169 to 10.885 mm. While the distance of the skin movement varied from one subject to another, a common skin movement direction was observed among the six subjects. For the repeatability evaluation of the proposed method, the MR images of the same subject were independently processed 10 times by a trained operator. The standard deviation of the skin movement on the second metacarpal of the index finger with respect to the predefined flexional motion ranged from 0.45 to 0.74 mm.
An artificial intelligence approach to registration of free-form shapes
2004, CIRP Annals - Manufacturing Technology
Registration, defined as the process of matching geometric entities, is performed when multiple scanned data sets must be aligned or when an existing model must match digitized point clouds. This process is crucial in several applications such as Reverse Engineering, CAD-based inspection and computer vision. The goal of this process is the computation of the optimal rigid transformation for the alignment of several sets of geometric entities (points and/or surfaces). Registration is generally performed by using a two-step procedure necessary to realize coarse and fine alignments. Human intervention is normally required for coarse registration while fine registration is usually a semi-automatic procedure. Consequently alignment is not usually a single step automatic operation and is also affect by errors.
In this paper the authors propose a hybrid approach for automatic registration applied to free-form shapes. This hybrid approach employs a asynchronous data communication between an Artificial Neural Network and Genetic Algorithms. The Neural Network performs the coarse alignment giving an initial solution for the registration operation which is then performed by Genetic Algorithms to minimize error deviations between geometrical entities. Several case studies have been investigated in order to validate the proposed approach.
Newton methods for parametric surface registration. Part I: Theory
2003, CAD Computer Aided Design
Citation Excerpt :
Today, the ICP method is widely used in registration applications. Currently surface registration research continues to expand on the ICP method by making additional enhancements or application specific modifications [4–9]. Free-form surface registration has a wide variety of applications including object recognition [10–12], autonomous vehicles for mining [13] and planetary rovers [14], medical imaging [15,16], computer-assisted and minimally invasive surgery [17,18], reverse engineering [19–21], and manufacturing inspection [22,23].
Surface registration is the process of determining the rigid body transformation that relocates a set of three-dimensional Cartesian coordinates as close as possible to a corresponding surface model. This procedure is used in free-form surface inspection, computer vision, and a wide variety of other applications. Newton methods for non-linear numerical minimization are rarely applied to this problem because of the high cost of computing the first and second derivatives through finite difference methods. This paper presents an approach to reduce this cost by providing analytic formulas for the derivatives. The resulting Newton methods are more efficient and accurate than those implemented in past research and have distinct advantages compared to the registration methods most widely used today. In this first part, the formulation of the analytic derivatives is presented along with a discussion of their application in various Newton methods.

View all citing articles on Scopus

View full text

Estimating pose uncertainty for surface registration

Abstract

Image and Vision Computing

Analysis of 3D rotation fitting

IEEE Transactions on Pattern Analysis and Machine Intelligence

A method for registration of 3D shapes

IEEE Transactions on Pattern Analysis and Machine Intelligence

Validation of 3D registration methods based on points and frames

Validation of 3D registration methods based on points and frames

INRIA Tech Report RR2470

Overview: computer vision performance characterization

Error propagation in machine vision

Machine Vision and Applications