DOCSLIB.ORG

Deep Probabilistic Regression of Elements of SO(3) Using Quaternion Averaging and Uncertainty Injection

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Deep Probabilistic Regression of Elements of SO(3) Using Quaternion Averaging and Uncertainty Injection Deep Probabilistic Regression of Elements of SO(3) using Quaternion Averaging and Uncertainty Injection Valentin Peretroukhin, Brandon Wagstaff, and Jonathan Kelly University of Toronto [email protected], [email protected], [email protected] Abstract HydraNet q1 <latexit sha1_base64="anqzf7dHa0EqnrCL0OfX9zzF89g=">AAACK3icjVDLSsNAFJ34tj6rSzeDRXBjSFRQdwU3bgQFq0JTys301g5OHp25kZaQ33CrP+DXuFLc+h9O0y5UXHhg4Mw598UJUyUNed6bMzU9Mzs3v7BYWVpeWV1br25cmyTTAhsiUYm+DcGgkjE2 SJLC21QjRKHCm/D+dOTfPKA2MomvaJhiK4K7WHalALJSEJwDaTnI+0Xbb6/XPPekBB+To8MJOfG573olamyCi3bVWQw6icgijEkoMKbpeym1ctAkhcKiEmQGUxD3cIdNS2OI0LTy8uiC71ilw7uJti8mXqrfO3KIjBlGoa2MgHrmtzcS//KaGXWPW7mM04wwFuNF3Uxx SvgoAd6RGgWpoSUgtLS3ctEDDYJsTj+2DFLQBgteCcppeT+TNss9BYSDPdNLNImMjGt/xf/Cu953/QPXuzys1b1JjAtsi22zXeazI1ZnZ+yCNZhgKXtkT+zZeXFenXfnY1w65Ux6NtkPOJ9fXsupUA==</latexit> SO(3) Consistent estimates of rotation are crucial to vision- q2 <latexit sha1_base64="X7QWZITLf+jirqY71Mk1DFiYv6Q=">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</latexit> <latexit sha1_base64="h3mNhBeZiXCMiG0iPSkfppXmRYA=">AAACK3icjVDLSsNAFJ34tj6rSzeDRXBjSFRQdwU3bgQFq0JTys301g5OHp25kZaQ33CrP+DXuFLc+h9O0y5UXHhg4Mw598UJUyUNed6bMzU9Mzs3v7BYWVpeWV1br25cmyTTAhsiUYm+DcGgkjE2 SJLC21QjRKHCm/D+dOTfPKA2MomvaJhiK4K7WHalALJSEJwDaTnI+0V7v71e89yTEnxMjg4n5MTnvuuVqLEJLtpVZzHoJCKLMCahwJim76XUykGTFAqLSpAZTEHcwx02LY0hQtPKy6MLvmOVDu8m2r6YeKl+78ghMmYYhbYyAuqZ395I/MtrZtQ9buUyTjPCWIwXdTPF KeGjBHhHahSkhpaA0NLeykUPNAiyOf3YMkhBGyx4JSin5f1M2iz3FBAO9kwv0SQyMq79Ff8L73rf9Q9c7/KwVvcmMS6wLbbNdpnPjlidnbEL1mCCpeyRPbFn58V5dd6dj3HplDPp2WQ/4Hx+AWCHqVE=</latexit> based motion estimation in augmented reality and robotics. … q, Σq <latexit sha1_base64="tZdxszg9zZNK/NhmJnp5g+eeSCw=">AAACRXicbZDPThsxEMa9QFsI/RPaIxerKVIPJdptkQCpB6ReekEC0QSkbBTNOpPEwmtv7FmUaLUvwNNwpS/QZ+hDcKt6bZ3NCpW2n2Tp8zcej/1LMiUdheH3YGV17dHjJ+sbjc2nz56/aG697DqT W4EdYZSxFwk4VFJjhyQpvMgsQpooPE8uPy3q51donTT6C80z7Kcw1nIkBZCPBs038TGCLuIuCjK2mJblOx4fA1k5K+IzOU6hHEwHzVbYDivxf01UmxardTLYCjbioRF5ipqEAud6UZhRvwBLUigsG3HuMANxCWPseashRdcvqu+UfMcnQz4y1i9NvEr/7CggdW6eJv5k CjRxf9cW4f9qvZxGB/1C6iwn1GI5aJQrToYv2PChtB6DmnsDwkr/Vi4mYEGQJ/hgyiwD67Dkjbi6rZjm0lPeVUA423UTY0nk5Np+V1bwDivxpdnfq81hdA+v+74dfWiHp3uto481xnW2zV6ztyxi++yIfWYnrMMEu2Y37JZ9Db4Fd8GP4Ofy6EpQ97xiDxT8+g2w9rNf </latexit> qh In this work, we present a method to extract probabilistic <latexit sha1_base64="KUVseSJeBDABueVivQQqnZmSEw4=">AAACK3icjVDLSsNAFJ34tj6rSzeDRXBjSFRQdwU3bgQFq0JTys301g5OHp25kZaQ33CrP+DXuFLc+h9O0y5UXHhg4Mw598UJUyUNed6bMzU9Mzs3v7BYWVpeWV1br25cmyTTAhsiUYm+DcGgkjE2 SJLC21QjRKHCm/D+dOTfPKA2MomvaJhiK4K7WHalALJSEJwDaTnI+0W7116vee5JCT4mR4cTcuJz3/VK1NgEF+2qsxh0EpFFGJNQYEzT91Jq5aBJCoVFJcgMpiDu4Q6blsYQoWnl5dEF37FKh3cTbV9MvFS/d+QQGTOMQlsZAfXMb28k/uU1M+oet3IZpxlhLMaLupni lPBRArwjNQpSQ0tAaGlv5aIHGgTZnH5sGaSgDRa8EpTT8n4mbZZ7CggHe6aXaBIZGdf+iv+Fd73v+geud3lYq3uTGBfYFttmu8xnR6zOztgFazDBUvbIntiz8+K8Ou/Ox7h0ypn0bLIfcD6/AL4vqYc=</latexit> estimates of rotation from deep regression models. First, Σ <latexit sha1_base64="fBF9yLlyYOeEEMmZny3OW4GfbXw=">AAACMHicjVDLSitBFOzxbfRq1KWbxiC4cZi5V1B3ghs3gqJRIRPCmc5J0tjzuN2nJWHIl7jVH/BrdCVu/Qo7kyxUXFjQUF11XlScK2koCF68qemZ2bn5hcXK0vKfldXq2vqVyawWWBeZyvRNDAaV TLFOkhTe5BohiRVex7fHI//6DrWRWXpJgxybCXRT2ZECyEmt6mp0CqRlv4guZDeBYataC/zDEnxM9vcm5DDkoR+UqLEJzlpr3mLUzoRNMCWhwJhGGOTULECTFAqHlcgazEHcQhcbjqaQoGkW5eVDvu2UNu9k2r2UeKl+7iggMWaQxK4yAeqZ795I/MlrWOocNAuZ5pYw FeNFHas4ZXwUA29LjYLUwBEQWrpbueiBBkEurC9b+jlog0NeicppxX8rXaC7Cgj7u6aXaRKWjO9+vwzv6q8f/vOD873aUTCJcYFtsi22w0K2z47YCTtjdSaYZffsgT16T96z9+q9jUunvEnPBvsC7/0Dm/Sq6w==</latexit> T?, Σ? we build on prior work and develop a multi-headed net- <latexit sha1_base64="NWYXWbGgVEsdc5qIXsxaq7W7mwQ=">AAACSHicbVDLbtNAFB0bCm2gkJYlmxERgkVj2SVSW6mLSmy6QSqiaSrFUXQ9uUlGGT+Yua4SWf4Evqbb9gf6B/0Ldqi7ThyDaOFII5055z5mTpQpacj3bx33ydO1Z8/XNxovXm6+et3c2j4zaa4F dkWqUn0egUElE+ySJIXnmUaII4W9aPZ56fcuUBuZJqe0yHAQwySRYymArDRsfgi/AGk5L07LYWgI9A7/rYTf5CSGWh42W77nV+D/kqAmLVbjZLjlbISjVOQxJiQUGNMP/IwGBWiSQmHZCHODGYgZTLBvaQIxmkFR/ajk760y4uNU25MQr9S/OwqIjVnEka2MgabmsbcU /+f1cxrvDwqZZDlhIlaLxrnilPJlPHwkNQpSC0tAaGnfysUUNAiyIT7YMs9AGyx5I6ymFd9zaYNuKyCct8001SRyMp69lVV4BxX4iux1anIQ/AnvbNcLPnn+107r6LCOcZ29Ze/YRxawPXbEjtkJ6zLBfrBLdsWunRvnp/PLuVuVuk7d84Y9gOveA1DUs6g=</latexi t> work structure we name HydraNet that can account for both Fused estimate Sensor data aleatoric and epistemic uncertainty. Second, we extend Hy- T, ΣT draNet to targets that belong to the rotation group, SO(3), <latexit sha1_base64="XoJsl6Acp9wI3OH7J/V/yM7Kucw=">AAACRXicbVBNT9tAEF1Dy0dKIcCxl1VDJQ4lsgEJkDggcekFiaoJIMVRNN5MklXWa7M7Roks/wF+DVf6B/ob+iN6q3ptN45blZYnrfTmvZmd3RelSlry/a/ewuKLl0vLK6u1V2uv1zfqm1tXNsmM wLZIVGJuIrCopMY2SVJ4kxqEOFJ4HY3PZ/71HRorE92iaYrdGIZaDqQAclKvvhNeIOg8vAAycpK3iuI9/12En+QwhqLX6tUbftMvwf8nQUUarMJlb9NbDfuJyGLUJBRY2wn8lLo5GJJCYVELM4spiDEMseOohhhtNy+/U/B3TunzQWLc0cRL9e+JHGJrp3HkOmOgkf3X m4nPeZ2MBsfdXOo0I9RivmiQKU4Jn2XD+9KgIDV1BISR7q1cjMCAIJfgky2TFIzFgtfC8rb8NpMu5T0FhJM9O0oMiYxs01VFGd5JCT4nR4cVOQn+hHe13wwOmv7Hw8bZaRXjCnvD3rJdFrAjdsY+sEvWZoLdswf2yD57X7xv3nfvx7x1watmttkTeD9/AU2Psyc=</la texit> SE(3) by regressing unit quaternions and using the tools of ro- Classical odometry <latexit sha1_base64="Hq8ax/xU8L8MmvyBsM+jfgLwLTU=">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</latexit> tation averaging and uncertainty injection onto the mani- fold to produce three-dimensional covariances. Finally, we Figure 1: Our model outputs a probabilistic representation present results and analysis on a synthetic dataset, learn of rotation as a unit quaternion and a full covariance ma- consistent orientation estimates on the 7-Scenes dataset, trix. A consistent uncertainty estimate allows us to improve and show how we can use our learned covariances to classical estimators using pose-graph optimization. fuse deep estimates of relative orientation with classical stereo visual odometry to improve localization on the KITTI less, the representational power of deep regression algo- dataset. rithms makes them an attractive option to complement clas- sical motion estimation when these latter methods perform poorly (e.g., under diverse lighting conditions or low scene 1. Introduction texture). By endowing deep regression models with a useful notion of uncertainty, we can account for out-of-training- Accounting for position and orientation, or pose, is at the distribution errors (a critical step to ensure safe operation) heart of many computer vision algorithms like visual odom- and fuse these models with classical methods using proba- etry, structure from motion, and SLAM. By tracking mo- bilistic factor graphs. In this work, we choose to focus on tion, these algorithms form the basis of visual localization rotation regression, since many motion algorithms are sen- pipelines in autonomous ground and aerial vehicles, visual sitive to rotation errors [18], and good rotation initializa- mapping, and augmented reality. tions can be critical to robust optimization [3]. Our novel Recent work [4, 15, 12] has attempted to transfer the suc- contributions are cess of deep neural networks in many areas of computer 1. a deep network structure we call HydraNet that builds vision to the task of camera pose estimation. These ap- on prior work [14, 17, 16] to build covariance matrices proaches, however, can produce arbitrarily poor pose esti- that account for both epistemic and aleatoric uncer- mates if sensor data differs from what is observed during tainty [13], training (i.e., it is ‘out of training distribution’) and their monolithic nature makes them difficult to debug. Further, 2. a formulation that extends HydraNet to means and co- despite much research effort, classical motion estimation variances defined on the rotation group SO(3), algorithms, like stereo visual odometry, still achieve state- of-the-art performance in nominal conditions1. Neverthe- 3. and open-source code for SO(3) regression2. 1Based on the KITTI odometry leaderboard [6] at the time of writing. 2https://github.com/utiasSTARS/so3_learning 432183 2. Approach where Log (·), represents the capitalized matrix logarithm · [19], and F the Frobenius norm. In the context of Equa- 2.1. HydraNet tion (1), using the angular metric leads to the Karcher mean, To endow deep networks with uncertainty, we extend which requires an iterative solver and has no known analytic prior work on multi-headed networks [17] and ensembles of expression. Applying the chordal metric leads to an analytic networks trained through bootstrapped training sets [14] to expression for the average but requires the use of Singular a single network structure we call HydraNet (see Figure 1). Value Decomposition. Using the quaternionic metric, how- HydraNet is composed of
Load more

Recommended publications
  • Molecular Symmetry
    Molecular Symmetry Symmetry helps us understand molecular structure, some chemical properties, and characteristics of physical properties (spectroscopy) – used with group theory to predict vibrational spectra for the identification of molecular shape, and as a tool for understanding electronic structure and bonding. Symmetrical : implies the species possesses a number of indistinguishable configurations. 1 Group Theory : mathematical treatment of symmetry. symmetry operation – an operation performed on an object which leaves it in a configuration that is indistinguishable from, and superimposable on, the original configuration. symmetry elements – the points, lines, or planes to which a symmetry operation is carried out. Element Operation Symbol Identity Identity E Symmetry plane Reflection in the plane σ Inversion center Inversion of a point x,y,z to -x,-y,-z i Proper axis Rotation by (360/n)° Cn 1. Rotation by (360/n)° Improper axis S 2. Reflection in plane perpendicular to rotation axis n Proper axes of rotation (C n) Rotation with respect to a line (axis of rotation). •Cn is a rotation of (360/n)°. •C2 = 180° rotation, C 3 = 120° rotation, C 4 = 90° rotation, C 5 = 72° rotation, C 6 = 60° rotation… •Each rotation brings you to an indistinguishable state from the original. However, rotation by 90° about the same axis does not give back the identical molecule. XeF 4 is square planar. Therefore H 2O does NOT possess It has four different C 2 axes. a C 4 symmetry axis. A C 4 axis out of the page is called the principle axis because it has the largest n . By convention, the principle axis is in the z-direction 2 3 Reflection through a planes of symmetry (mirror plane) If reflection of all parts of a molecule through a plane produced an indistinguishable configuration, the symmetry element is called a mirror plane or plane of symmetry .
    [Show full text]
  • (II) I Main Topics a Rotations Using a Stereonet B
    GG303 Lecture 11 9/4/01 1 ROTATIONS (II) I Main Topics A Rotations using a stereonet B Comments on rotations using stereonets and matrices C Uses of rotation in geology (and engineering) II II Rotations using a stereonet A Best uses 1 Rotation axis is vertical 2 Rotation axis is horizontal (e.g., to restore tilted beds) B Construction technique 1 Find orientation of rotation axis 2 Find angle of rotation and rotate pole to plane (or a linear feature) along a small circle perpendicular to the rotation axis. a For a horizontal rotation axis the small circle is vertical b For a vertical rotation axis the small circle is horizontal 3 WARNING: DON'T rotate a plane by rotating its dip direction vector; this doesn't work (see handout) IIIComments on rotations using stereonets and matrices Advantages Disadvantages Stereonets Good for visualization Relatively slow Can bring into field Need good stereonets, paper Matrices Speed and flexibility Computer really required to Good for multiple rotations cut down on errors A General comments about stereonets vs. rotation matrices 1 With stereonets, an object is usually considered to be rotated and the coordinate axes are held fixed. 2 With rotation matrices, an object is usually considered to be held fixed and the coordinate axes are rotated. B To return tilted bedding to horizontal choose a rotation axis that coincides with the direction of strike. The angle of rotation is the negative of the dip of the bedding (right-hand rule! ) if the bedding is to be rotated back to horizontal and positive if the axes are to be rotated to the plane of the tilted bedding.
    [Show full text]
  • Euler Quaternions
    Four different ways to represent rotation my head is spinning... The space of rotations SO( 3) = {R ∈ R3×3 | RRT = I,det(R) = +1} Special orthogonal group(3): Why det( R ) = ± 1 ? − = − Rotations preserve distance: Rp1 Rp2 p1 p2 Rotations preserve orientation: ( ) × ( ) = ( × ) Rp1 Rp2 R p1 p2 The space of rotations SO( 3) = {R ∈ R3×3 | RRT = I,det(R) = +1} Special orthogonal group(3): Why it’s a group: • Closed under multiplication: if ∈ ( ) then ∈ ( ) R1, R2 SO 3 R1R2 SO 3 • Has an identity: ∃ ∈ ( ) = I SO 3 s.t. IR1 R1 • Has a unique inverse… • Is associative… Why orthogonal: • vectors in matrix are orthogonal Why it’s special: det( R ) = + 1 , NOT det(R) = ±1 Right hand coordinate system Possible rotation representations You need at least three numbers to represent an arbitrary rotation in SO(3) (Euler theorem). Some three-number representations: • ZYZ Euler angles • ZYX Euler angles (roll, pitch, yaw) • Axis angle One four-number representation: • quaternions ZYZ Euler Angles φ = θ rzyz ψ φ − φ cos sin 0 To get from A to B: φ = φ φ Rz ( ) sin cos 0 1. Rotate φ about z axis 0 0 1 θ θ 2. Then rotate θ about y axis cos 0 sin θ = ψ Ry ( ) 0 1 0 3. Then rotate about z axis − sinθ 0 cosθ ψ − ψ cos sin 0 ψ = ψ ψ Rz ( ) sin cos 0 0 0 1 ZYZ Euler Angles φ θ ψ Remember that R z ( ) R y ( ) R z ( ) encode the desired rotation in the pre- rotation reference frame: φ = pre−rotation Rz ( ) Rpost−rotation Therefore, the sequence of rotations is concatentated as follows: (φ θ ψ ) = φ θ ψ Rzyz , , Rz ( )Ry ( )Rz ( ) φ − φ θ θ ψ − ψ cos sin 0 cos 0 sin cos sin 0 (φ θ ψ ) = φ φ ψ ψ Rzyz , , sin cos 0 0 1 0 sin cos 0 0 0 1− sinθ 0 cosθ 0 0 1 − − − cφ cθ cψ sφ sψ cφ cθ sψ sφ cψ cφ sθ (φ θ ψ ) = + − + Rzyz , , sφ cθ cψ cφ sψ sφ cθ sψ cφ cψ sφ sθ − sθ cψ sθ sψ cθ ZYX Euler Angles (roll, pitch, yaw) φ − φ cos sin 0 To get from A to B: φ = φ φ Rz ( ) sin cos 0 1.
    [Show full text]
  • The Structuring of Pattern Repeats in the Paracas Necropolis Embroideries
    University of Nebraska - Lincoln DigitalCommons@University of Nebraska - Lincoln Textile Society of America Symposium Proceedings Textile Society of America 1988 Orientation And Symmetry: The Structuring Of Pattern Repeats In The Paracas Necropolis Embroideries Mary Frame University of British Columbia Follow this and additional works at: https://digitalcommons.unl.edu/tsaconf Part of the Art and Design Commons Frame, Mary, "Orientation And Symmetry: The Structuring Of Pattern Repeats In The Paracas Necropolis Embroideries" (1988). Textile Society of America Symposium Proceedings. 637. https://digitalcommons.unl.edu/tsaconf/637 This Article is brought to you for free and open access by the Textile Society of America at DigitalCommons@University of Nebraska - Lincoln. It has been accepted for inclusion in Textile Society of America Symposium Proceedings by an authorized administrator of DigitalCommons@University of Nebraska - Lincoln. 136 RESEARCH-IN-PROGRESS REPORT ORIENTATION AND SYMMETRY: THE STRUCTURING OF PATTERN REPEATS IN THE PARACAS NECROPOLIS EMBROIDERIES MARY FRAME RESEARCH ASSOCIATE, INSTITUTE OF ANDEAN STUDIES 4611 Marine Drive West Vancouver, B.C. V7W 2P1 The most extensive Peruvian fabric remains come from the archeological site of Paracas Necropolis on the South Coast of Peru. Preserved by the dry desert conditions, this cache of 429 mummy bundles, excavated in 1926-27, provides an unparal- leled opportunity for comparing the range and nature of varia- tions in similar fabrics which are securely related in time and space. The bundles are thought to span the time period from 500-200 B.C. The most numerous and notable fabrics are embroideries: garments that have been classified as mantles, tunics, wrap- around skirts, loincloths, turbans and ponchos.
    [Show full text]
  • Rotation Matrix - Wikipedia, the Free Encyclopedia Page 1 of 22
    Rotation matrix - Wikipedia, the free encyclopedia Page 1 of 22 Rotation matrix From Wikipedia, the free encyclopedia In linear algebra, a rotation matrix is a matrix that is used to perform a rotation in Euclidean space. For example the matrix rotates points in the xy -Cartesian plane counterclockwise through an angle θ about the origin of the Cartesian coordinate system. To perform the rotation, the position of each point must be represented by a column vector v, containing the coordinates of the point. A rotated vector is obtained by using the matrix multiplication Rv (see below for details). In two and three dimensions, rotation matrices are among the simplest algebraic descriptions of rotations, and are used extensively for computations in geometry, physics, and computer graphics. Though most applications involve rotations in two or three dimensions, rotation matrices can be defined for n-dimensional space. Rotation matrices are always square, with real entries. Algebraically, a rotation matrix in n-dimensions is a n × n special orthogonal matrix, i.e. an orthogonal matrix whose determinant is 1: . The set of all rotation matrices forms a group, known as the rotation group or the special orthogonal group. It is a subset of the orthogonal group, which includes reflections and consists of all orthogonal matrices with determinant 1 or -1, and of the special linear group, which includes all volume-preserving transformations and consists of matrices with determinant 1. Contents 1 Rotations in two dimensions 1.1 Non-standard orientation
    [Show full text]
  • The Problem of Exterior Orientation in Photogrammetry
    T he Problem of Exterior Orientation in Photogrammetry*t GEORGE H. ROSENFIELD, Data Reduction A nalyst, RCA Service Co., Air Force Missile Test Center, Patrick Air Force Base, Fla. ABSTRACT: Increased interest in Analytical Photogrammetry has focused at­ tention upon a chaotic situation which complicates research and communication between photogrammetrists. The designation of the orientation parameters in a different manner by almost every investigator has resulted in the need for a universal non-ambiguous system which depicts the actual physical orientation of the photograph. This paper presents a discussion of the basic concepts in­ volved in the various systems in use, and recommends the adoption of a standard system within the basic framework of the Stockholm resolution of July 1956, concerning the sign convention to be used in Photogrammetry. 1. INTRODUCTION CHAOTIC situation exists within the science of Photogrammetry which complicates A research and communication between photogrammetrists. When the theoretician studies a technical paper, he finds the work unnecessarily laborious until he has first become familiar with the coordinate system, definitions, and rotations employed by the writer. \i\Then the skilled instrument operator sits down to a newstereoplotting instrument he finds the orienta­ tion procedure unnecessarily difficult until he be­ comes familiar with the positive direction of the translational and rotational motions. Thus it is that almost every individual has devised his own unique system of designating the exterior orienta­ tion of the photograph. In order to alleviate these complications, the VIII International Congress for Photogrammetry at Stockholm in July 1956, accepted a resolution on the sign convention to be used in photogrammetry.
    [Show full text]
  • A Guided Tour to the Plane-Based Geometric Algebra PGA
    A Guided Tour to the Plane-Based Geometric Algebra PGA Leo Dorst University of Amsterdam Version 1.15{ July 6, 2020 Planes are the primitive elements for the constructions of objects and oper- ators in Euclidean geometry. Triangulated meshes are built from them, and reflections in multiple planes are a mathematically pure way to construct Euclidean motions. A geometric algebra based on planes is therefore a natural choice to unify objects and operators for Euclidean geometry. The usual claims of `com- pleteness' of the GA approach leads us to hope that it might contain, in a single framework, all representations ever designed for Euclidean geometry - including normal vectors, directions as points at infinity, Pl¨ucker coordinates for lines, quaternions as 3D rotations around the origin, and dual quaternions for rigid body motions; and even spinors. This text provides a guided tour to this algebra of planes PGA. It indeed shows how all such computationally efficient methods are incorporated and related. We will see how the PGA elements naturally group into blocks of four coordinates in an implementation, and how this more complete under- standing of the embedding suggests some handy choices to avoid extraneous computations. In the unified PGA framework, one never switches between efficient representations for subtasks, and this obviously saves any time spent on data conversions. Relative to other treatments of PGA, this text is rather light on the mathematics. Where you see careful derivations, they involve the aspects of orientation and magnitude. These features have been neglected by authors focussing on the mathematical beauty of the projective nature of the algebra.
    [Show full text]
  • Euclidean Space - Wikipedia, the Free Encyclopedia Page 1 of 5
    Euclidean space - Wikipedia, the free encyclopedia Page 1 of 5 Euclidean space From Wikipedia, the free encyclopedia In mathematics, Euclidean space is the Euclidean plane and three-dimensional space of Euclidean geometry, as well as the generalizations of these notions to higher dimensions. The term “Euclidean” distinguishes these spaces from the curved spaces of non-Euclidean geometry and Einstein's general theory of relativity, and is named for the Greek mathematician Euclid of Alexandria. Classical Greek geometry defined the Euclidean plane and Euclidean three-dimensional space using certain postulates, while the other properties of these spaces were deduced as theorems. In modern mathematics, it is more common to define Euclidean space using Cartesian coordinates and the ideas of analytic geometry. This approach brings the tools of algebra and calculus to bear on questions of geometry, and Every point in three-dimensional has the advantage that it generalizes easily to Euclidean Euclidean space is determined by three spaces of more than three dimensions. coordinates. From the modern viewpoint, there is essentially only one Euclidean space of each dimension. In dimension one this is the real line; in dimension two it is the Cartesian plane; and in higher dimensions it is the real coordinate space with three or more real number coordinates. Thus a point in Euclidean space is a tuple of real numbers, and distances are defined using the Euclidean distance formula. Mathematicians often denote the n-dimensional Euclidean space by , or sometimes if they wish to emphasize its Euclidean nature. Euclidean spaces have finite dimension. Contents 1 Intuitive overview 2 Real coordinate space 3 Euclidean structure 4 Topology of Euclidean space 5 Generalizations 6 See also 7 References Intuitive overview One way to think of the Euclidean plane is as a set of points satisfying certain relationships, expressible in terms of distance and angle.
    [Show full text]
  • A Tutorial on Euler Angles and Quaternions
    A Tutorial on Euler Angles and Quaternions Moti Ben-Ari Department of Science Teaching Weizmann Institute of Science http://www.weizmann.ac.il/sci-tea/benari/ Version 2.0.1 c 2014–17 by Moti Ben-Ari. This work is licensed under the Creative Commons Attribution-ShareAlike 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/ by-sa/3.0/ or send a letter to Creative Commons, 444 Castro Street, Suite 900, Mountain View, California, 94041, USA. Chapter 1 Introduction You sitting in an airplane at night, watching a movie displayed on the screen attached to the seat in front of you. The airplane gently banks to the left. You may feel the slight acceleration, but you won’t see any change in the position of the movie screen. Both you and the screen are in the same frame of reference, so unless you stand up or make another move, the position and orientation of the screen relative to your position and orientation won’t change. The same is not true with respect to your position and orientation relative to the frame of reference of the earth. The airplane is moving forward at a very high speed and the bank changes your orientation with respect to the earth. The transformation of coordinates (position and orientation) from one frame of reference is a fundamental operation in several areas: flight control of aircraft and rockets, move- ment of manipulators in robotics, and computer graphics. This tutorial introduces the mathematics of rotations using two formalisms: (1) Euler angles are the angles of rotation of a three-dimensional coordinate frame.
    [Show full text]
  • Rigid Motion – a Transformation That Preserves Distance and Angle Measure (The Shapes Are Congruent, Angles Are Congruent)
    REVIEW OF TRANSFORMATIONS GEOMETRY Rigid motion – A transformation that preserves distance and angle measure (the shapes are congruent, angles are congruent). Isometry – A transformation that preserves distance (the shapes are congruent). Direct Isometry – A transformation that preserves orientation, the order of the lettering (ABCD) in the figure and the image are the same, either both clockwise or both counterclockwise (the shapes are congruent). Opposite Isometry – A transformation that DOES NOT preserve orientation, the order of the lettering (ABCD) is reversed, either clockwise or counterclockwise becomes clockwise (the shapes are congruent). Composition of transformations – is a combination of 2 or more transformations. In a composition, you perform each transformation on the image of the preceding transformation. Example: rRx axis O,180 , the little circle tells us that this is a composition of transformations, we also execute the transformations from right to left (backwards). If you would like a visual of this information or if you would like to quiz yourself, go to http://www.mathsisfun.com/geometry/transformations.html. Line Reflection Point Reflection Translations (Shift) Rotations (Turn) Glide Reflection Dilations (Multiply) Rigid Motion Rigid Motion Rigid Motion Rigid Motion Rigid Motion Not a rigid motion Opposite isometry Direct isometry Direct isometry Direct isometry Opposite isometry NOT an isometry Reverse orientation Same orientation Same orientation Same orientation Reverse orientation Same orientation Properties preserved: Properties preserved: Properties preserved: Properties preserved: Properties preserved: Properties preserved: 1. Distance 1. Distance 1. Distance 1. Distance 1. Distance 1. Angle measure 2. Angle measure 2. Angle measure 2. Angle measure 2. Angle measure 2. Angle measure 2.
    [Show full text]
  • Working with 3D Rotations
    Working With 3D Rotations Stan Melax Graphics Software Engineer, Intel Human Brain is wired for Spatial Computation Which shape is the same: a) b) c) “I don’t need to ask A childhood IQ test question for directions” Translations Rotations Agenda ● Rotations and Matrices (hopefully review) ● Combining Rotations ● Matrix and Axis Angle ● Challenges of deep Space (of Rotations) ● Quaternions ● Applications Terminology Clarification Preferred usages of various terms: Linear Angular Object Pose Position (point) Orientation A change in Pose Translation (vector) Rotation Rate of change Linear Velocity Spin also: Direction specifies 2 DOF, Orientation specifies all 3 angular DOF. Rotations Trickier than Translations Translations Rotations a then b == b then a x then y != y then x (non-commutative) ● Programming with rotations also more challenging! 2D Rotation θ Rotate [1 0] by θ about origin 1,1 [ cos(θ) sin(θ) ] θ θ sin cos θ 1,0 2D Rotation θ Rotate [0 1] by θ about origin -1,1 0,1 sin θ [-sin(θ) cos(θ)] θ θ cos 2D Rotation of an arbitrary point Rotate about origin by θ = cos θ + sin θ 2D Rotation of an arbitrary point 푥 Rotate 푦 about origin by θ 푥′, 푦′ 푥′ = 푥 cos θ − 푦 sin θ 푦′ = 푥 sin θ + 푦 cos θ 푦 푥 2D Rotation Matrix 푥 Rotate 푦 about origin by θ 푥′, 푦′ 푥′ = 푥 cos θ − 푦 sin θ 푦′ = 푥 sin θ + 푦 cos θ 푦 푥′ cos θ − sin θ 푥 푥 = 푦′ sin θ cos θ 푦 cos θ − sin θ Matrix is rotation by θ sin θ cos θ 2D Orientation 풚 Yellow grid placed over first grid but at angle of θ cos θ − sin θ sin θ cos θ 풙 Columns of the matrix are the directions of the axes.
    [Show full text]
  • Closed-Form Solution of Absolute Orientation Using Unit Quaternions
    Berthold K. P. Horn Vol. 4, No. 4/April 1987/J. Opt. Soc. Am. A 629 Closed-form solution of absolute orientation using unit quaternions Berthold K. P. Horn Department of Electrical Engineering, University of Hawaii at Manoa, Honolulu, Hawaii 96720 Received August 6, 1986; accepted November 25, 1986 Finding the relationship between two coordinate systems using pairs of measurements of the coordinates of a number of points in both systems is a classic photogrammetric task. It finds applications in stereophotogrammetry and in robotics. I present here a closed-form solution to the least-squares problem for three or more points. Currently various empirical, graphical, and numerical iterative methods are in use. Derivation of the solution is simplified by use of unit quaternions to represent rotation. I emphasize a symmetry property that a solution to this problem ought to possess. The best translational offset is the difference between the centroid of the coordinates in one system and the rotated and scaled centroid of the coordinates in the other system. The best scale is equal to the ratio of the root-mean-square deviations of the coordinates in the two systems from their respective centroids. These exact results are to be preferred to approximate methods based on measurements of a few selected points. The unit quaternion representing the best rotation is the eigenvector associated with the most positive eigenvalue of a symmetric 4 X 4 matrix. The elements of this matrix are combinations of sums of products of corresponding coordinates of the points. 1. INTRODUCTION I present a closed-form solution to the least-squares prob- lem in Sections 2 and 4 and show in Section 5 that it simpli- Suppose that we are given the coordinates of a number of fies greatly when only three points are used.
    [Show full text]