Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Parameter estimation class 5 Multiple View Geometry Comp 290-089 Marc Pollefeys Content • Background: Projective geometry (2D, 3D), Parameter estimation, Algorithm evaluation. • Single View: Camera model, Calibration, Single View Geometry. • Two Views: Epipolar Geometry, 3D reconstruction, Computing F, Computing structure, Plane and homographies. • Three Views: Trifocal Tensor, Computing T. • More Views: N-Linearities, Multiple view reconstruction, Bundle adjustment, autocalibration, Dynamic SfM, Cheirality, Duality Multiple View Geometry course schedule (subject to change) Jan. 7, 9 Intro & motivation Projective 2D Geometry Jan. 14, 16 (no class) Projective 2D Geometry Jan. 21, 23 Projective 3D Geometry (no class) Jan. 28, 30 Parameter Estimation Parameter Estimation Feb. 4, 6 Algorithm Evaluation Camera Models Feb. 11, 13 Camera Calibration Single View Geometry Feb. 18, 20 Epipolar Geometry 3D reconstruction Feb. 25, 27 Fund. Matrix Comp. Structure Comp. Planes & Homographies Trifocal Tensor Mar. 18, 20 Three View Reconstruction Multiple View Geometry Mar. 25, 27 MultipleView Reconstruction Bundle adjustment Apr. 1, 3 Auto-Calibration Papers Apr. 8, 10 Dynamic SfM Papers Apr. 15, 17 Cheirality Papers Apr. 22, 24 Duality Project Demos Mar. 4, 6 Projective 3D Geometry • Points, lines, planes and quadrics • Transformations • П∞, ω∞ and Ω ∞ Singular Value Decomposition A UΣ V T UΣ Σ VT X Homogeneous least-squares min AX subject to X 1 solution X Vn Parameter estimation • 2D homography Given a set of (xi,xi’), compute H (xi’=Hxi) • 3D to 2D camera projection Given a set of (Xi,xi), compute P (xi=PXi) • Fundamental matrix Given a set of (xi,xi’), compute F (xi’TFxi=0) • Trifocal tensor Given a set of (xi,xi’,xi”), compute T Number of measurements required • At least as many independent equations as degrees of freedom required • Example: x h11 h12 h λ yx' h Hx 21 22 1 h31 h32 h13 x h23 y h33 1 2 independent equations / point 8 degrees of freedom 4x2≥8 Approximate solutions • Minimal solution 4 points yield an exact solution for H • More points • No exact solution, because measurements are inexact (“noise”) • Search for “best” according to some cost function • Algebraic or geometric/statistical cost Gold Standard algorithm • Cost function that is optimal for some assumptions • Computational algorithm that minimizes it is called “Gold Standard” algorithm • Other algorithms can then be compared to it Direct Linear Transformation (DLT) xxii Hx i 0 h 1T x T i T xi xi, yi, wi Hx i h 2 x i 3T yh 3T x wh 2 T x h x i i i i T i 1 3T x i Hx i wih x i xih x i 2T 1T xih x i yih x i 0T T wi x i yix iT T wi x i T 1 yi x i h 2 T T 0 xix i h 0 T T 3 xi x i 0 h Ai h 0 • Direct Linear Transformation (DLT) Equations are linear in h Ai h 0 • Only 2 out of 3 are linearly independent (indeed, 2 eq/pt) 0TT wix TiT 0 T wTix i 0T wwixxi T 0 T T i i yix i xix i 1 2 yix TiT h1 yix i T 2 xix Ti h 0 xiTx i 3 0 h 3 xi A wi Aifi w0i’≠0) (only drop i ythird i A i row • Holds for any homogeneous representation, e.g. (xi’,yi’,1) Direct Linear Transformation (DLT) • Solving for H A1 A 2Ah h 0 A 3 or 12x9, but rank 8 size A is 8x9 A 4 Trivial solution is h=09T is not interesting 1-D null-space yields solution of interest pick for example the one with h 1 Direct Linear Transformation (DLT) • Over-determined solution A1 A 2Ah h 0 of inexact measurement No exact solution because An i.e. “noise” Find approximate solution - Additional constraint needed to avoid 0, e.g. - Ah 0 not possible, so minimize Ah h 1 DLT algorithm Objective Given n≥4 2D to 2D point correspondences {xi↔xi’}, determine the 2D homography matrix H such that xi’=Hxi Algorithm (i) For each correspondence xi ↔xi’ compute Ai. Usually only two first rows needed. (ii) Assemble n 2x9 matrices Ai into a single 2nx9 matrix A (iii) Obtain SVD of A. Solution for h is last column of V (iv) Determine H from h Inhomogeneous solution Since h can only be computed up to scale, ~ pick hj=1, e.g. h9=1, and solve for 8-vector h 0 0 xi wi ' yi wi ' wi wi ' xi yi ' yi yi ' ~ wi yi ' 0 h x w ' y w ' w w ' 0 0 0 xi xi ' yi xi ' i i i i i i wi xi ' Solve using Gaussian elimination (4 points) or using linear least-squares (more than 4 points) However, if h9=0 this approach fails also poor results if h9 close to zero Therefore, not recommended Note h9=H33=0 if origin is mapped to infinity 0 lT Hx 0 0 0 1H 0 0 1 Degenerate configurations x1 x4 x2 x3 Constraints: Define: Then, x1 H? x4 x2 x3 (case A) xi Hx i 0 H’? (case B) x1 x4 x2 x3 i=1,2,3,4 H* x4lT H* x i x4 lT x i 0, i 1,2,3 H*x 4 x4 lT x 4 kx4 H* is rank-1 matrix and thus not a homography If H* is unique solution, then no homography mapping xi→xi’(case B) If further solution H exist, then also αH*+βH (case A) (2-D null-space in stead of 1-D null-space) Solutions from lines, etc. 2D homographies from 2D lines li H T li Ah 0 Minimum of 4 lines 3D Homographies (15 dof) Minimum of 5 points or 5 planes 2D affinities (6 dof) Minimum of 3 points or lines Conic provides 5 constraints Mixed configurations? Cost functions • Algebraic distance • Geometric distance • Reprojection error • Comparison • Geometric interpretation • Sampson error Algebraic distance DLT minimizes e Ah ei Ah residual vector partial vector for each (xi↔xi’) algebraic error vector 0 wix d alg xi , Hx i ei T T w x 0 i i algebraic distance T 2 2 T i yix h xix T i T i 2 d alg x1 , x 2 a12 a22 where a a1 , a2 , a3 T x1 x 2 2 2 d x , Hx alg i i ei i 2 Ah e 2 2 i Not geometrically/statistically meaningfull, but given good normalization it works fine and is very fast (use for initialization) Geometric distance x measured coordinates x̂ estimated coordinates x true coordinates d(.,.) Euclidean distance (in image) Error in one image 2 e.g. calibration pattern Ĥ argmin d xi , Hx i H i Symmetric transfer error Ĥ argmin H d x i , H xi d xi , Hx i -1 2 2 i Reprojection error Ĥ, x̂ , x̂ argmin d x , x̂ d x , x̂ 2 i i H,x̂ i , x̂ i i i i 2 i i subject to x̂i Ĥx̂ i Reprojection error d x, H x d x, Hx -1 2 2 d x, x̂ d x, x̂ 2 2 Comparison of geometric and algebraic distances Error in one image T T ˆ ˆ ˆ x i xi , yi , wi x̂ i xi , yi , wi Hx 1 h T T yw Tw y ˆ ˆ 0 wixi i yi ix i i i 2 h A i hT ei T T 0 wi xˆixixxi iwˆi 3 wix i h d alg xi , x̂i yiwˆ i wi yˆ i wixˆi xiwˆ i 2 2 2 d xi , x̂i yi / wi yˆ i / wˆ i xˆi / wˆ i xi / wi d alg xi , x̂i / wiwˆ i 2 2 2 1/ 2 wi 1 typical, but not wˆ i h 3 x i, except for affinities For affinities DLT can minimize geometric distance Possibility for iterative algorithm Geometric interpretation of reprojection error Estimating homography~fit surface ν Hto points X=(x,y,x’,y’)T in 4 xi Hx i 0 represents 2 quadrics in 4 (quadratic in X) X i X̂ i 2 2 2 2 2 xi xˆi yi yˆ i xi xˆi yi yˆ i 2 2 d x i , x̂ i d xi , x̂i d x i , x̂ i d xi , x̂i d Xi , H 2 2 2 Analog to conic fitting d alg x, C x T Cx 2 d x , C 2 Sampson error between algebraic and geometric error 2 Vector X̂ that minimizes the geometric error X X̂ is the closest point on the variety ν H to the measurement X Sampson error: 1st order approximation of X̂ Ah CH X 0 CH CH X δ X CH X δX X C CH X H δ X 0 X δ X X̂ X CH X̂ 0 CH JδX e with J X Find the vector δ X that minimizes δ X subject to JδX e Find the vector δ X that minimizes δ X subject to JδX e Use Lagrange multipliers: minimize δ X δ X - 2λJδ X e 0 T derivatives 2δ X - 2λ J 0 T T 2JδX e 0 δX JT λ JJ T λ e 0 e J JJ λ JJ T δX X̂ X δ X δX 2 T δTXδX eT JJ T 1 e 1 T 1 e Sampson error between algebraic and geometric error 2 Vector X̂ that minimizes the geometric error X X̂ is the closest point on the variety ν H to the measurement X Sampson error: 1st order approximation of X̂ Ah CH X 0 CH CH X δ X CH X δX X C CH X H δ X 0 X δ X X̂ X CH X̂ 0 JδX e Find the vector δ X that minimizes δ X subject to JδX e δX 2 δ δ e JJ T X X T T 1 e (Sampson error) Sampson approximation δX 2 e JJ T T 1 e A few points (i) (ii) (iii) For a 2D homography X=(x,y,x’,y’) e CH X is the algebraic error vector CH is a 2x4 matrix, J X e.g. J wx T h 2 yx T h 3 / x wh yh 11 i i i i i 21 i 31 2 T (iv) Similar to algebraic error e e e (v) 2 in fact, same as Mahalanobis distance e T JJ Sampson error independent of linear reparametrization (cancels out in between e and J) T T 1 e JJ e (vi) Must be summed for all points (vii) Close to geometric error, but much fewer unknowns Statistical cost function and Maximum Likelihood Estimation • Optimal cost function related to noise model • Assume zero-mean isotropic Gaussian noise (assume outliers removed) Pr x 1 2 πσ 2 e d x, x 2 / 2 2 Error in one image Pr xi | H i 1 2 πσ log Pr xi | H xi ,Hx i 2 / 2 2 e 2 d 1 2σ 2 2 d x i , Hx i constant Maximum Likelihood Estimate 2 d x , H x i i Statistical cost function and Maximum Likelihood Estimation • Optimal cost function related to noise model • Assume zero-mean isotropic Gaussian noise (assume outliers removed) Pr x 1 2 πσ 2 e d x, x 2 / 2 2 Error in both images Prxi | H i 1 2πσ 2 e d x i , x i 2 d Maximum Likelihood Estimate d x , x̂ 2 i i 2 d x i , x̂ i xi ,Hxi 2 / 2 2 Mahalanobis distance • General Gaussian case Measurement X with covariance matrix Σ X X X X 1 X X 2 T Error in two images (independent) 2 2 X X X X Varying covariances X i i Xi 2 i Xi Xi 2 i Next class: Parameter estimation (continued) Transformation invariance and normalization Iterative minimization Robust estimation Upcoming assignment • Take two or more photographs taken from a single viewpoint • Compute panorama • Use different measures DLT, MLE • Use Matlab • Due Feb. 13