Home

ECCV 2012 tutorial Computational photography

Multi-View Geometry and Computational Photography using Non-Classical Cameras

Amit Agrawal and Srikumar Ramalingam

October 7, 2012 · Half-day morning session

Overview

Model cameras that see from more than one viewpoint.

A perspective camera captures a scene from a single viewpoint. A multi-perspective camera combines rays from several viewpoints into one image. This tutorial provides a practical guide to designing, modeling, and calibrating non-classical cameras for wide-angle imaging, 3D reconstruction, light-field capture, novel-view rendering, and digital refocusing.

ECCV 2012 conference banner
01

Course topics

The tutorial follows the development of a complete imaging system, from capture and physical modeling to calibration, synthesis, and reconstruction.

  • Camera designs: omnidirectional cameras, non-central catadioptric cameras using mirrors, refractive-sphere cameras, and light-field cameras using masks and lenslets.
  • Geometry and calibration: analytical forward-projection models, calibration algorithms, and the reconstruction artifacts introduced by central-camera approximations.
  • Algebraic tools: Gröbner bases and related methods for formulating and solving equations arising from non-classical camera geometry.
02

Course slides

03

Tutorial videos