260 likes | 270 Views
Dive into the world of linear algebra as it applies to 3D vision. Explore transformations, orthogonalization, eigenvalues, symmetric and skew-symmetric matrices, SVD, and optimization techniques.
E N D
Lecture 5 Introduction to Linear Algebra Shankar Sastry September 13th, 2005 Invitation to 3D vision
Orthogonal group • What is the set of transformations that preserve the inner product? • Remember inner product under a transformation? • More on this later … Invitation to 3D vision
Gram-Schmidt orthogonalization MEMENTO! will appear in calibration (aka Q-R) Structure of the Parameter matrix Invitation to 3D vision
Structure induced by a linear map A X X’ Ra(A) T T Ra(A ) Nu(A) T T Nu(A ) Ra(A) Nu(A) Invitation to 3D vision
Eigenvalues and eigenvectors • Eigenvalues and eigenvectors encode the “essence” of the linear map represented by A: the range space, the null space, the rank, the norm etc. • How do the notions of eigenvalues and eigenvectors generalize to NON-SQUARE matrices? • SVD, later … Invitation to 3D vision
Symmetric matrices Invitation to 3D vision
Symmetric matrices (contd.) Invitation to 3D vision
Skew-symmetric matrices Invitation to 3D vision
Skew-symmetric matrices (contd.) Invitation to 3D vision
The singular value decomposition Invitation to 3D vision
The SVD (contd.) Invitation to 3D vision
The SVD: geometric interpretation A Invitation to 3D vision
Pseudo-inverse and linear systems Invitation to 3D vision
Fixed-rank approximation • Useful for matrix factorization • MEMENTO! Invitation to 3D vision
Transformation groups Invitation to 3D vision
Affine transformation • Not a linear transformation! • Can be made linear in HOMOGENEOUS COORDINATES MEMENTO! will appear everywhere Invitation to 3D vision
Affine group (contd.) • Composition of affine transformations. • What is the inverse transformation? Invitation to 3D vision
Orthogonal group • What is the set of transformations that preserve the inner product? • Remember inner product under a transformation? • More on this later … Invitation to 3D vision
Euclidean group Invitation to 3D vision
Unconstrained optimization Invitation to 3D vision
Unconstrained optimization (contd.) Invitation to 3D vision
Iterative minimization (local) • Steepest descent: • Newton’s method: • More in general: Invitation to 3D vision
Gauss-Newton, Levemberg-Marquardt • Quadratic cost function • No second derivatives Invitation to 3D vision
Constrained optimization Invitation to 3D vision
Lagrangian function and multipliers Invitation to 3D vision