8. Multiple-View Geometry for Image-Based Modeling
Sunday, Full Day, 8:30 am - 5:30 pm
Room 501AB
Level: Intermediate

A comprehensive introduction to multiple-view geometry, with a coherent spectrum of algorithms that systematically extract 3D information (motion, structure, and camera calibration) from 2D perspective images. The technical core of the course provides a simple characterization of all constraints among multiple images that can be utilized for 3D reconstruction in a simple matrix rank framework. The course covers essentially all the technical steps and details necessary for the entire reconstruction process: image formation, feature extraction, feature tracking and matching, two-view reconstruction, camera calibration and auto-calibration, multiple-view reconstruction, and incorporation of various types of scene knowledge. It provides a step-by-step recipe for 3D reconstruction from image sequences and demonstrates a real-time 3D motion and structure estimation system, together with its application to real-time virtual insertion in live video. The course also incorporates the most recent developments on reconstruction of dynamic scenes containing multiple moving objects.

Applications of this technology range from special effects (scene or camera motion capture, virtual insertion, image-based modeling and rendering) to computer graphics, photogrammetry, surveillance, autonomous robotics, medical imaging, and virtual reality.

Prerequisites
Basic knowledge of image formation and modeling, 2D and 3D geometry, rigid-body transformations, and linear algebra, as obtainable from experience with image-based rendering or an introductory-level computer vision or graphics course. Unlike traditional treatment of this subject, this course does not require knowledge of tensorial algebra or projective geometry.

Intended Audience
Academicians, graduate students, industrial researchers or developers who are interested in state-of-the-art vision theory and techniques for applications such as 3D modeling, motion capture, visualization, virtual insertion and reality, and video indexing and editing.

Organizer
Yi Ma
University of Illinois at Urbana-Champaign

Lecturers
Jana Kosecka
George Mason University

Yi Ma
University of Illinois at Urbana-Champaign

Stefano Soatto
University of California, Los Angeles

Rene Vidal
John Hopkins University

Schedule

	Component 1: Image Formation and Primitives
8:30	Introduction to Multiple-View,Geometry,Imaging Geometry and Image, Formation Image Primitives and Correspondence Soatto
9:45	Vision Applications/Design Rensink
10:15	Break
	Component 2: Two-View Geometry
10:30	Reconstruction From Two Calibrated Views Kosecka
12:15	Lunch
	Component 3: Multiple-View Geometry
1:45	Geometry and Algebra of Multiple Views Vidal
2:35	Multiple-view Reconstruction From Scene Knowledge Ma
3:30	Break
	Component 4: Image-Based Modeling and Motion Segmentation
3:45	Step-by-Step 3D Modeling From Images Kosecka and Stefano
4:30	Motion and Image Segmentation Vidal
5:10	Summary, Questions, and Answers All