Multiview Geometry in Computer Vision

This site provides information on the CS 8803D course with the title above, to be taught in the Fall 2002 Semester. The site will be updated frequently, as the course moves on. So stay tuned!

In this course, we will explore the secrets of how to build 3D models from camera images or video footage, how stereo-vision works, why you cannot go to a point at infinity but you can nevertheless see it, how we can put synthetic objects into existing movies and have them look real, etc... All this is really fun stuff and we will do this in a hands-on way, with lots of little MATLAB scripts rather than cumbersome C++ behemoths. Our journey into the strange and wonderful world of multiview geometry will be mediated by the recent and very nice book 'Multiple View Geometry', by Richard Hartley and Andrew Zisserman, which should be available from the bookstore.

The course should be quite appealing to anybody interested in vision, robotics, augmented reality, special effects, and graphics. It will probably be taught only every other year, depending on demand.

Announcements will be posted on the git.cc.class.cs8803d newsgroup

In these pages you’ll find more information on course logistics, the schedule, grading and policy, assignments, and related links.

Frank Dellaert