Broad Area Colloquium For AI-Geometry-Graphics-Robotics-Vision
(CS 528)
Visual 3D modeling of real-world objects, scenes and events from videos
Marc Pollefeys
Department of Computer Science
University of North Carolina
April 30, 2007, 4:15PM
TCSeq 200
http://graphics.stanford.edu/ba-colloquium/
Abstract
Images and videos form a rich source of information about the visual
world. The extraction of 3D information from images is an important
research problem in computer vision and graphics. The ubiquitous
presence of cameras and the tremendous advances of processing and
communication technologies yields important opportunities and challenges
in those areas.
My work has focused on developing flexible techniques for recovering 3D
shape, motion and appearance from images. A first example of this is an
approach to recover photo-realistic 3D models of objects or scenes from
videos. A key aspect of our approach is the ability to also recover the
geometric and photometric calibration of the camera from the image data
so that our techniques can also work with uncalibrated consumer cameras
or archive photographs. Our recent work in this area has focussed on
real-time reconstruction of urban environments from video streams.
Besides static scenes, we also aim to model dynamic events. While it
can be possible to recover dynamic shape from a single video stream, the
simplest way to recover dynamic shapes consists of using multiple
cameras. Here also the complete calibration and synchronization can be
obtained from video recorded during normal operation. I will present a
shape reconstruction approaches that recovers the 3D shape of dynamic
objects as well as static occluding objects in the environment.
Towards the future, one of my main research goals is to develop
approaches for capturing immersive 4D spatio-temporal representation of
dynamic events taking place in large-scale environments.