Question? Leave a message!




Coordinates and Transformations

Coordinates and Transformations
Coordinates and Transformations MIT ECCS 6.837 Wojciech Matusik many slides follow Steven Gortler’s book 1 Hierarchical modeling • Many coordinate systems: • Camera • Static scene • car • driver • arm • hand Image courtesy of Gunnar A. Sjögren on Wikimedia Commons. License: CC-BY-SA. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. • ... • Makes it important to understand coordinate systems 2 Coordinates • We are used to represent points with tuples of coordinates such as • But the tuples are meaningless without a clear coordinate system could be this point could be this point in the red in the blue coordinate system coordinate system 3 Different objects • Points • represent locations • Vectors • represent movement, force, displacement from A to B • Normals • represent orientation, unit length • Coordinates • numerical representation of the above objects in a given coordinate system 4 Points & vectors are different • The 0 vector has a fundamental meaning: no movement, no force • Why would there be a special 0 point? • It’s meaningful to add vectors, not points • Boston location + NYC location =? + =? 5 Points & vectors are different • Moving car • points describe location of car elements • vectors describe velocity, distance between pairs of points • If I translate the moving car to a different road • The points (location) change • The vectors (speed, distance between points) don’t Image courtesy of Gunnar A. Sjögren on Wikimedia Commons. License: CC-BY-SA. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. 6 Matrices have two purposes • (At least for geometry) • Transform things • e.g. rotate the car from facing North to facing East • Express coordinate system changes • e.g. given the driver's location in the coordinate system of the car, express it in the coordinate system of the world 7 Goals for today • Make it very explicit what coordinate system is used • Understand how to change coordinate systems • Understand how to transform objects • Understand difference between points, vectors, normals and their coordinates 8 Questions? 9 Reference • This lecture follows the new book by Steven (Shlomo) Gortler from Harvard: Foundations of 3D Computer Graphics 10 Plan • Vectors • Points • Homogeneous coordinates • Normals (in the next lecture) 11 Vectors (linear space) • Formally, a set of elements equipped with addition and scalar multiplication • plus other nice properties • There is a special element, the zero vector • no displacement, no force 12 Vectors (linear space) • We can use a basis to produce all the vectors in the space: • Given n basis vectors any vector can be written as here: 13 Linear algebra notation • can be written as • Nice because it makes the basis (coordinate system) explicit • Shorthand: • where bold means triplet, t is transpose 14 Questions? 15 Linear transformation Courtesy of Prof. Fredo Durand. Used with permission. • Transformation of the vector space 16 Linear transformation Courtesy of Prof. Fredo Durand. Used with permission. • Transformation of the vector space so that • Note that it implies • Notation for transformations 17 Matrix notation • Linearity implies ? 18 Matrix notation • Linearity implies • i.e. we only need to know the basis transformation • or in algebra notation 19 Algebra notation • The are also vectors of the space • They can be expressed in the basis for example: ... • which gives us 20 Algebra notation • The are also vectors of the space • They can be expressed in the basis for example: • which gives us 21 Recap, matrix notation • Given the coordinates c in basis the transformed vector has coordinates Mc in 22 Why do we care • We like linear algebra • It’s always good to get back to an abstraction that we know and for which smarter people have developed a lot of tools • But we also need to keep track of what basis/coordinate system we use 23 Questions? 24 Change of basis • Critical in computer graphics • From world to car to arm to hand coordinate system • From Bezier splines to B splines and back • problem with basis change: you never remember which is M or M¯¹ it’s hard to keep track of where you are 25 Change of basis • Assume we have two bases and • And we have the coordinates of in • e.g. • i.e. • which implies 26 Change of basis • We have & • Given the coordinate of in : • What are the coordinates in ? 27 Change of basis • We have & • Given the coordinate of in : • Replace by its expression in • has coordinates in • Note how we keep track of the coordinate system by having the basis on the left 28 Questions? 29 Linear Transformations •L(p + q) = L(p) + L(q) •L(ap) = a L(p) Similitudes Translation is not linear: Linear Rigid / Euclidean f(p) = p+t Scaling Identity Translation Isotropic Scaling Reflection f(ap) = ap+t ≠ a(p+t) = a f(p) Rotation Shear f(p+q) = p+q+t ≠ (p+t)+(q+t) = f(p) + f(q) 30 Plan • Vectors • Points • Homogenous coordinates • Normals 31 Points vs. Vectors • A point is a location • A vector is a motion between two points • Adding vectors is meaningful • going 3km North + 4km East = going 5km North-East • Adding points is not meaningful • Boston location + New York location = ? • Multiplying a point by a scalar? • The zero vector is meaningful (no movement) • Zero point ? 32 Affine space • Points are elements of an affine space • We denote them with a tilde • Affine spaces are an extension of vector spaces 33 Point-vector operations • Subtracting points gives a vector • Adding a vector to a point gives a point 34 Frames • A frame is an origin plus a basis • We can obtain any point in the space by adding a vector to the origin • using the coordinates c of the vector in 35 Algebra notation • We like matrix-vector expressions • We want to keep track of the frame • We’re going to cheat a little for elegance and decide that 1 times a point is the point • is represented in by 4 coordinate, where the extra dummy coordinate is always 1 (for now) 36 Recap • Vectors can be expressed in a basis • Keep track of basis with left notation • Change basis • Points can be expressed in a frame (origin+basis) • Keep track of frame with left notation • adds a dummy 4th coordinate always 1 37 Affine transformations • Include all linear transformations • Applied to the vector basis • Plus translation Courtesy of Prof. Fredo Durand. Used with permission. 38 Matrix notation • We know how to transform the vector basis • We will soon add translation by a vector 39 Linear component • Note how we leave the fourth component alone 40 Translation component • Express translation vector t in the basis 41 Translation 42 Full affine expression Which tells us both how to get a new frame ftM or how to get the coordinates Mc after transformation 43 Questions? 44 More notation properties • If the fourth coordinate is zero, we get a vector • Subtracting two points: • Gives us a vector (last coordinate = 0) 45 More notation properties • Adding a point to a vector • Gives us a point (4th coordinate=1) 46 More notation properties • vectors are not affected by the translation part • because their 4th coordinate is 0 • If I rotate my moving car in the world, I want its motion to rotate • If I translate it, motion should be unaffected 47 Questions? 48 Frames & hierarchical modeling • Many coordinate systems (frames): • Camera • Static scene • car • driver • arm • hand • ... Image courtesy of Gunnar A. Sjögren on Wikimedia Commons. License: CC-BY-SA. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. • Need to understand nested transformations 49 Frames & hierarchical modeling • Example: what if I rotate the wheel of the moving car: • frame 1: world • frame 2: car • transformation: rotation Image courtesy of Gunnar A. Sjögren on Wikimedia Commons. License: CC-BY-SA. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. 50 Frames & transformations • Transformation S wrt car frame f • how is the world frame a affected by this? • we have • which gives • i.e. the transformation in a is A-1SA • i.e., from right to left, A takes us from a to f, then we apply S, then we go back to a with A-1 51 Questions? 52 How are transforms combined? Scale then Translate (5,3) (2,2) Scale(2,2) Translate(3,1) (1,1) (3,1) (0,0) (0,0) Use matrix multiplication: p' = T ( S p ) = TS p 0 0 0 1 3 2 0 2 3 1 2 2 0 1 0 0 0 1 TS = = 0 0 1 0 0 1 0 0 1 Caution: matrix multiplication is NOT commutative 53 53 Non-commutative Composition Scale then Translate: p' = T ( S p ) = TS p (5,3) (2,2) Scale(2,2) Translate(3,1) (1,1) (3,1) (0,0) (0,0) Translate then Scale: p' = S ( T p ) = ST p (8,4) (4,2) Translate(3,1) Scale(2,2) (6,2) (1,1) (3,1) (0,0) 54 Non-commutative Composition Scale then Translate: p' = T ( S p ) = TS p 0 0 0 1 3 2 0 2 3 1 2 2 0 1 0 0 0 1 TS = = 0 0 0 0 1 0 1 0 1 Translate then Scale: p' = S ( T p ) = ST p 0 0 0 2 0 1 3 2 6 2 1 2 0 0 0 1 0 2 ST = = 0 0 1 0 0 1 0 0 1 55 Questions? 56 Plan • Vectors • Points • Homogenous coordinates • Normals 57 Forward reference and eye • The fourth coordinate is useful for perspective projection • Called homogenous coordinates 58 Homogeneous Coordinates •Add an extra dimension (same as frames) • in 2D, we use 3-vectors and 3 x 3 matrices • In 3D, we use 4-vectors and 4 x 4 matrices •The extra coordinate is now an arbitrary value, w • You can think of it as “scale,” or “weight” • For all transformations except perspective, you can c x x' a b just set w=1 and not worry = y‘ d e f y about it 1 1 1 0 0 59 Projective Equivalence • All non-zero scalar multiples of a point are considered identical • to get the equivalent Euclidean point, divide by w ax x/w x w =0 ay y/w y az z/w = z = aw 1 w a = 0 60 Why bother with extra coord? • This picture gives away almost the whole story. w = 1 w = 2 61 Perspective in 2D • Camera at origin, looking along z, 90 degree f.o.v., “image plane” at z=1 This image is in the public domain. Source: http://openclipart.org/detail/34051/digicam-by-thesaurus. 62 Perspective in 2D The projected point in homogeneous coordinates (we just added w=1): This image is in the public domain. Source: http://openclipart.org/detail/34051/digicam-by-thesaurus. 63 Perspective in 2D Projectively equivalent This image is in the public domain. Source: http://openclipart.org/detail/34051/digicam-by-thesaurus. 64 Perspective in 2D We’ll just copy z to w, and get the projected point after homogenization This image is in the public domain. Source: http://openclipart.org/detail/34051/digicam-by-thesaurus. 65 Homogeneous Visualization • Divide by w to normalize (project) (0,0,0) (0, 0, 1) = (0, 0, 2) = … w = 1 (7, 1, 1) = (14, 2, 2) = … (4, 5, 1) = (8, 10, 2) = … w = 2 66 Homogeneous Visualization • Divide by w to normalize (project) Points at infinity (directions) • w = 0? (0,0,0) (0, 0, 1) = (0, 0, 2) = … w = 1 (7, 1, 1) = (14, 2, 2) = … (4, 5, 1) = (8, 10, 2) = … w = 2 67 Projective Equivalence – Why? • For affine transformations, adding w=1 in the end proved to be convenient. • The real showpiece is perspective. © source unknown. All rights reserved. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-us e/. This image is in the public domain. Source: http://openclipart.org/detail/34051/digicam-by-thesaurus. 68 Questions? 69 Eye candy: photo tourism • Application of homogenous coordinates • Goal: given N photos of a scene • find where they were taken • get 3D geometry for points in the scene ©ACM. All rights reserved. This content is excluded from our Creative Commons license. For more information, see http://ocw.mit.edu/help/faq-fair-use/. From Photo Tourism:: Exploring Photo Collections in 3D, used with permission from ACM, Inc. 70 Step 1: point correspondences • Extract salient points (corners) from images • Find the same scene point in other images • To learn how it’s done, take 6.815 71 Structure from motion • Given point correspondences • Unknowns: 3D point location, camera poses • For each point in each image, write perspective equations Minimize f(R,T,P) p1 Camera 3 R3,t3 Camera 1 R1,t1 Camera 2 R2,t2 72 Eye candy: photo tourism QuickTime™ and a MPEG-4 Video decompressor are needed to see this picture. 73 And that’s it for today • The rest on Thursday 74 Normal • Surface Normal: unit vector that is locally perpendicular to the surface 75 Why is the Normal important? • It's used for shading — makes things look 3D Diffuse Shading object color only 76 Visualization of Surface Normal ± x = Red ± y = Green ± z = Blue 77 How do we transform normals? nWS nOS World Space Object Space 78 Transform Normal like Object? •translation? •rotation? •isotropic scale? •scale? •reflection? •shear? •perspective? 79 Transform Normal like Object? •translation? •rotation? •isotropic scale? •scale? •reflection? •shear? •perspective? 80 Transformation for shear and scale Incorrect Normal Transformation Correct Normal Transformation 81 More Normal Visualizations Incorrect Normal Transformation Correct Normal Transformation 82 So how do we do it right? •Think about transforming the tangent plane to the normal, not the normal vector nOS nWS vWS vOS Original Incorrect Correct Pick any vector vOS in the tangent plane, how is it transformed by matrix M? vWS = M vOS 83 Transform tangent vector v v is perpendicular to normal n: Dot product nOSᵀ vOS = 0 nOS nOS ᵀ (Mˉ¹ M) vOS = 0 (nOSᵀ Mˉ¹) (M vOS) = 0 vOS (nOSᵀ Mˉ¹) vWS = 0 vWS is perpendicular to normal nWS: nWSᵀ = nOSᵀ (Mˉ¹) nWS nWS = (Mˉ¹)ᵀ nOS vWS nWSᵀ vWS = 0 84 Digression nWS = (Mˉ¹)ᵀ nOS • The previous proof is not quite rigorous; first you’d need to prove that tangents indeed transform with M. • Turns out they do, but we’ll take it on faith here. • If you believe that, then the above formula follows. 85 Comment • So the correct way to transform normals is: nWS = (Mˉ¹)ᵀ nOS Sometimes denoted Mˉᵀ • But why did nWS = M nOS work for similitudes? • Because for similitude / similarity transforms, (Mˉ¹)ᵀ =λ M • e.g. for orthonormal basis: Mˉ¹ = M ᵀ i.e. (Mˉ¹)ᵀ = M 86 Connections • Not part of class, but cool • “Covariant”: transformed by the matrix • e.g., tangent • “Contravariant”: transformed by the inverse transpose • e.g., the normal • a normal is a “co-vector” • Google “differential geometry” to find out more 87 Questions? 88 That’s All for Today • Further Reading –Buss, Chapter 2 • Other Cool Stuff –Algebraic Groups –http://phototour.cs.washington.edu/ –http://phototour.cs.washington.edu/findingpaths/ –Free-form deformation of solid objects –Harmonic coordinates for character articulation 89 MIT OpenCourseWare http://ocw.mit.edu 6.837 Computer Graphics Fall 2012 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
Website URL
Comment
sharer
Presentations
Free
Document Information
Category:
Presentations
User Name:
Dr.ShaneMatts
User Type:
Teacher
Country:
United States
Uploaded Date:
23-07-2017