54: Obtaining 3D Models With a Hand-Held Camera

Wednesday, Tutorial
10:30 am – 12:15 pm
Room 006CD

How 3D models can be obtained from images acquired with a hand-held camera. The approach is based on advanced automatic techniques that avoid camera calibration and a priori scene knowledge, which gradually retrieve more and more information about the images, the cameras, and the scene.

Basic understanding of 3D geometry and perspective projection. Full appreciation of this course requires some knowledge of projective geometry, computer vision, and image-processing techniques.

Feature-point extraction and matching, multi-view relations, robust matching, projective structure and motion recovery, self-calibration, bundle adjustment, image-pair rectification, dense stereo matching, multi-view matching, 3D surface modeling, texturing, and applications.

Marc Pollefeys
Katholieke Universiteit Leuven


10:30 Introduction
10:45 Structure and Motion Recovery
-- Feature Point Extraction and Matching
-- Relating Images
-- Projective Structure and Motion
-- Self-Calibration
11:45 Dense Model Reconstruction
-- Dense Depth Estimation
-- Modeling
12:15 Examples, Applications and Questions

Additional information from SIGGRAPH 2002 Courses is available in the Course Notes

Back To Main Page
conference exhibition Call for Participation Presenters Media Registration Search SIGGRAPH 2002 Homepage