MIT Department of Electrical Engineering & Computer Science

EECS Spring 2000 Catalogue Supplement

6.965 Multiple Views Geometry: An Approach to Three-Dimensional Computer Vision

TR 9:30-11, 4-145
Prof. Olivier Faugeras, NE43-713, x8-8826
Prerequisite: basic notions of linear algebra, calculus, probabilities, random processes, image and signal processing.
3-0-9

NOTE: This subject will be offered 28 February - 17 March and 27 March - 14 April

The course covers intermediate and advanced topics in three-dimensional Computer Vision. It is intended to provide the theoretical tools that are necessary to tackle important applications of Computer Vision such as the interactive and automatic modelling of three-dimensional objects and scenes, the navigation of robots, image based rendering, augmented reality. Twelve lectures will be given over six weeks. Lecture notes based on a MIT Press book entitled "Multiple Views Geometry" by the author and Q.-T. Luong will be handed out.

PLAN

I) Projective Geometry

II) Grassmann-Cayley algebra

III) Modelling cameras - Perspective, paraperspective, affine models, nonlinear distortions

IV) Systems of two cameras: epipolar geometry, Fundamental matrixes, the role of planes, stereo

V) Practical estimation of the Fundamental matrix, outlier rejection

VI) Stratification of binocular vision: projective, affine and Euclidean settings. Applications: obstacle detection, image mosaics, image based rendering

VII) Systems of three cameras: the Trifocal tensor

VIII) Practical estimation of the Trifocal tensor

IX) Stratification of n >=3 views: projective, affine and Euclidean settings. Applications: 3-D modelling, augmented reality

X) Self-calibration of systems of cameras