Computer Vision - Teoria (2008/2009)

Learning outcomes

The course aims at providing the student with the theoretical and
practical tools to tackle the problem of recovering the 3D
structure of a scene starting from its 2D projections, the
images. The focus is on the geometry of the problem. The methods
will be discussed in sufficient detail to allow the student to
implement them on a computer.

* Introduction, course overview, motivations.
* Geometry of the pin-hole camera.
* Camera calibration (or resection) with DLT.
* Epipolar geometry: Essential matrix, Fundamental matrix.
* Estimating the Fundamental matrix with DLT
* Triangulation.
* Planes and homographies: infinite plane homography, mosaics, parallax.
* Estimating the homography with DLT.
* 3-D Reconstruction from 2 views: up to a similarity, up to a projective transform.
* Trifocal geometry: trifocal constraints for points and lines.
* Reconstruction from multiple views: Euclidean, projective.
* Multifocal constraints.
* Autocalibration: direct method, stratification, solution strategies.
* Dealing with noisy data: pre-conditioning, algebraic vs geometric errors.
* Robust estimation: M-estimators, RANSAC, LMedS.
* Practical calibration, radial distortion.
* Epioolar rectification.
* Absolute orientation (with scaling), exterior orientation.

Oral.