An integrated solution for reducing ill-conditioning and testing the results in non-linear 3D similarity transformations
04 May 2018-Inverse Problems in Science and Engineering (Taylor & Francis)-Vol. 26, Iss: 5, pp 708-727
TL;DR: In this article, a combined solution is proposed to reduce ill-conditioning and to perform precision analysis and global outlier test in linearized least squares (LLS) and direct (non-iterative) least squares.
Abstract: To find 3D similarity transformation parameters (a scale, three rotational angles, and three translation elements) between two orthogonal coordinate systems in 3D is an ill-posed non-linear inverse problem by means of common points (their Cartesian components are known in the both systems). The problem can be solved via Linearized Least Squares (LLS) or Direct (non-iterative) Least Squares (DLS). Since the parameters in LLS take different quantities (and units) from each other, the condition problems can arise during the solution of normal equations. In this paper, we propose a combined solution to reducing ill-conditioning and to perform precision analysis and global outlier test in LLS accordingly. The way is based on column norming and uses the normalized unknowns instead of the original ones at the solution stage of the normal equations. While the global outlier test is fulfilled on the normalized unknowns, the original unknowns and their precisions obtained using the normalized matrix with li...
Citations
More filters
TL;DR: In this paper, the authors presented a rigid iterative algorithm of Helmert transformation using a unit dual quaternion and showed that the accuracy of computed parameter is comparable to the classic Procrustes algorithm from Grafarend and Awange.
Abstract: The rigid motion involving both rotation and translation in the 3D space can be simultaneously described by a unit dual quaternion. Considering this excellent property, the paper constructs the Helmert transformation (seven-parameter similarity transformation) model based on a unit dual quaternion and then presents a rigid iterative algorithm of Helmert transformation using a unit dual quaternion. Because of the singularity of the coefficient matrix of the normal equation, the nine parameter (including one scale factor and eight parameters of a dual quaternion) Helmert transformation model is reduced into five parameter (including one scale factor and four parameters of a unit quaternion which can represent the rotation matrix) Helmert transformation one. Besides, a good start estimate of parameter is required for the iterative algorithm, hence another algorithm employed to compute the initial value of parameter is put forward. The numerical experiments involving a case of small rotation angles i.e. geodetic coordinate transformation and a case of big rotation angles i.e. the registration of LIDAR points are studied. The results show the presented algorithms in this paper are correct and valid for the two cases, disregarding the rotation angles are big or small. And the accuracy of computed parameter is comparable to the classic Procrustes algorithm from Grafarend and Awange (J Geod 77:66–76, 2003), the orthonormal matrix algorithm from Zeng (Earth Planets Space 67:105, 2015), and the algorithm from Wang et al. (J Photogramm Remote Sens 94:63–69, 2014).
10 citations
TL;DR: A weighted total least squares (WTLS) iterative algorithm of the 3D similarity coordinate transformation based on Gibbs vectors is proposed that is fast in terms of fewer iterations, reliable and does not need good initial values of transformation parameters.
Abstract: The 3D similarity coordinate transformation is fundamental and frequently encountered in many areas of work such as geodesy, engineering surveying, LIDAR, terrestrial laser scanning, photogrammetry, machine vision, etc. The algorithms of 3D similarity transformation are divided into two categories. One is a closed-form algorithm that is straightforward and fast. However, it cannot provide the accuracy information for the transformation parameters. The other category of algorithm is iterative, and this can offer the accuracy information for the transformation parameters. However, the latter usually needs a good initial value of the unknown. Considering the accuracy information for transformation parameters is essential or indispensable from the viewpoint of uncertainty, this contribution proposes a weighted total least squares (WTLS) iterative algorithm of the 3D similarity coordinate transformation based on Gibbs vectors. It is fast in terms of fewer iterations, reliable and does not need good initial values of transformation parameters. Two cases including the registration of LIDAR points with big rotation angles and a geodetic datum transformation with small rotation angles are demonstrated to validate the new algorithm.
8 citations
Additional excerpts
...2016; Kurt 2018)....
[...]
...(2016), Kurt (2018), Zeng et al....
[...]
TL;DR: Eigenvalue-based approaches, i.e., Tikhonov regularization using eigenvalue and Moore–Penrose (pseudo) inversing, provide the best numerical solutions among the approaches initially investigated by the authors.
Abstract: Sensor orientation is an essential step for georeferencing of images, establishing the coordinate transformation between the image and ground spaces. This orientation is mainly carried out by the sensor dependent or independent models, and the desired georeferencing accuracy is achieved with the adjusted orientation elements using ancillary data. Independently of the orientation model, an ill-posed problem is occurred in the adjustment process, caused by the ill-conditioned Jacobian matrix which is differentiated through the orientation elements. However, this challenge is mitigated by various approaches such as regularization, matrix inversion, or elimination methods. In this work, three different types of orientation models were exposed on Zonguldak test site characterizing mountainous urban and dense forest areas: 1) sensor-dependent orientation model for handling of SPOT 5 HRG panchromatic stereo images and 2) sensor-dependent and 3) sensor-independent rational function model (RFM) for Pleiades 1A and SPOT 6 panchromatic mono images. The main finding of this work is that eigenvalue-based approaches, i.e., Tikhonov regularization using eigenvalue and Moore–Penrose (pseudo) inversing, provide the best numerical solutions among the approaches initially investigated by the authors. The novelty of this article is that two eigenvalue-based approaches investigated in this article are never preferred to solve ill-posed problem in different sensor orientation cases.
2 citations
Cites background from "An integrated solution for reducing..."
...In the case of cond(A) < 10−2, the matrix is assumed as ill-conditioned [37]....
[...]
TL;DR: In this article, the authors proposed an extended weighted total least squares (WTLS) iterative algorithm of 3D similarity transformation based on Gibbs vector, which treats the transformation parameters and the target coordinate of non-control points as unknowns.
Abstract: Considering coordinate errors of both control points and non-control points, and different weights between control points and non-control points, this contribution proposes an extended weighted total least squares (WTLS) iterative algorithm of 3D similarity transformation based on Gibbs vector. It treats the transformation parameters and the target coordinate of non-control points as unknowns. Thus it is able to recover the transformation parameters and compute the target coordinate of non-control points simultaneously. It is also able to assess the accuracy of the transformation parameters and the target coordinates of non-control points. Obviously it is different from the traditional algorithms that first recover the transformation parameters and then compute the target coordinate of non-control points by the estimated transformation parameters. Besides it utilizes a Gibbs vector to represent the rotation matrix. This representation does not introduce additional unknowns; neither introduces transcendental function like sine or cosine functions. As a result, the presented algorithm is not dependent to the initial value of transformation parameters. This excellent performance ensures the presented algorithm is suitable for the big rotation angles. Two numerical cases with big rotation angles including a real world case (LIDAR point cloud registration) and a simulative case are tested to validate the presented algorithm.
2 citations
TL;DR: In this paper , the point-wise weighted 3D coordinate transformation using a unit dual quaternion is formulated and derived in detail, and four numerical cases, including geodetic datum transformation, registration of LIDAR point clouds, and two simulated cases, are studied.
Abstract: Abstract Considering that a unit dual quaternion can describe elegantly the rigid transformation including rotation and translation, the point-wise weighted 3D coordinate transformation using a unit dual quaternion is formulated. The constructed transformation model by a unit dual quaternion does not need differential process to eliminate the three translation parameters, while traditional models do. Based on the Lagrangian extremum law, the analytical dual quaternion algorithm (ADQA) of the point-wise weighted 3D coordinate transformation is proved existed and derived in detail. Four numerical cases, including geodetic datum transformation, the registration of LIDAR point clouds, and two simulated cases, are studied. This study shows that ADQA is valid as well as the modified procrustes algorithm (MPA) and the orthonormal matrix algorithm (OMA). ADQA is suitable for the 3D coordinate transformation with point-wise weight and no matter rotation angles are small or big. In addition, the results also indicate that if the distribution of common points degrades from 3D or 2D space to 1D space, the solvable correct transformation parameters decrease. In other words, all common points should not be located on a line. From the perspective of improving the transformation accuracy, high accurate control points (with small errors in the coordinates) should be chosen, and it is preferred to decrease the rotation angles as much as possible. Graphical Abstract
References
More filters
11,285 citations
TL;DR: Anyone involved in scientific computing ought to have a copy of at least one version of Numerical Recipes, and there also ought to be copies in every library.
Abstract: The two Numerical Recipes books are marvellous. The principal book, The Art of Scientific Computing, contains program listings for almost every conceivable requirement, and it also contains a well written discussion of the algorithms and the numerical methods involved. The Example Book provides a complete driving program, with helpful notes, for nearly all the routines in the principal book. The first edition of Numerical Recipes: The Art of Scientific Computing was published in 1986 in two versions, one with programs in Fortran, the other with programs in Pascal. There were subsequent versions with programs in BASIC and in C. The second, enlarged edition was published in 1992, again in two versions, one with programs in Fortran (NR(F)), the other with programs in C (NR(C)). In 1996 the authors produced Numerical Recipes in Fortran 90: The Art of Parallel Scientific Computing as a supplement, called Volume 2, with the original (Fortran) version referred to as Volume 1. Numerical Recipes in C++ (NR(C++)) is another version of the 1992 edition. The numerical recipes are also available on a CD ROM: if you want to use any of the recipes, I would strongly advise you to buy the CD ROM. The CD ROM contains the programs in all the languages. When the first edition was published I bought it, and have also bought copies of the other editions as they have appeared. Anyone involved in scientific computing ought to have a copy of at least one version of Numerical Recipes, and there also ought to be copies in every library. If you already have NR(F), should you buy the NR(C++) and, if not, which version should you buy? In the preface to Volume 2 of NR(F), the authors say 'C and C++ programmers have not been far from our minds as we have written this volume, and we think that you will find that time spent in absorbing its principal lessons will be amply repaid in the future as C and C++ eventually develop standard parallel extensions'. In the preface and introduction to NR(C++), the authors point out some of the problems in the use of C++ in scientific computing. I have not found any mention of parallel computing in NR(C++). Fortran has quite a lot going for it. As someone who has used it in most of its versions from Fortran II, I have seen it develop and leave behind other languages promoted by various enthusiasts: who now uses Algol or Pascal? I think it unlikely that C++ will disappear: it was devised as a systems language, and can also be used for other purposes such as scientific computing. It is possible that Fortran will disappear, but Fortran has the strengths that it can develop, that there are extensive Fortran subroutine libraries, and that it has been developed for parallel computing. To argue with programmers as to which is the best language to use is sterile. If you wish to use C++, then buy NR(C++), but you should also look at volume 2 of NR(F). If you are a Fortran programmer, then make sure you have NR(F), volumes 1 and 2. But whichever language you use, make sure you have one version or the other, and the CD ROM. The Example Book provides listings of complete programs to run nearly all the routines in NR, frequently based on cases where an anlytical solution is available. It is helpful when developing a new program incorporating an unfamiliar routine to see that routine actually working, and this is what the programs in the Example Book achieve. I started teaching computational physics before Numerical Recipes was published. If I were starting again, I would make heavy use of both The Art of Scientific Computing and of the Example Book. Every computational physics teaching laboratory should have both volumes: the programs in the Example Book are included on the CD ROM, but the extra commentary in the book itself is of considerable value. P Borcherds
1,367 citations
Book•
01 Jan 1988TL;DR: This textbook on theoretical geodesy deals with the estimation of unknown parameters, the testing of hypothesis and the estimationof intervals in linear models and most of the necessary theorems of vector and matrix-algebra and the probability distributions for the test statistics are derived.
Abstract: This textbook on theoretical geodesy deals with the estimation of unknown parameters, the testing of hypothesis and the estimation of intervals in linear models. The reader will find presentations of the Gauss-Markoff model, the analysis of variance, the multivariate model, the model with unknown variance and covariance components and the regression model, as well as the mixed model for estimation random parameters. To make the book self-contained most of the necessary theorems of vector and matrix-algebra and the probability distributions for the test statistics are derived. Students of geodesy, as well as of mathematics and engineering, will find the geodetical application of mathematical and statistical models extremely useful.
1,200 citations
"An integrated solution for reducing..." refers background in this paper
...For testing procedure in detail, the reader can take a look in [25]....
[...]
Book•
01 Jan 2008
TL;DR: The next generation of positioning models for positioning and data processing will depend on the design of the satellite itself, as well as on the satellite orbits it is placed in.
Abstract: Reference systems.- Satellite orbits.- Satellite signals.- Observables.- Mathematical models for positioning.- Data processing.- Data transformation.- GPS.- Glonass.- Galileo.- More on GNSS.- Applications.- Conclusion and outlook.
893 citations
Book•
01 Mar 2011
TL;DR: This book is intended primarily as a handbook for engineers who must design practical systems so that the reader may design an estimator that meets all application requirements and is robust to modeling assumptions.
Abstract: This book is intended primarily as a handbook for engineers who must design practical systems. Its primary goal is to discuss model development in sufficient detail so that the reader may design an estimator that meets all application requirements and is robust to modeling assumptions. Since it is sometimes difficult to a priori determine the best model structure, use of exploratory data analysis to define model structure is discussed. Methods for deciding on the “best” model are also presented.
220 citations