3D Structure from 2D Motion

Tony Jebara, Ali Azarbayejani and Alex Pentland
MIT Media Laboratory, Cambridge MA, 02139
{ jebara, ali, sandy }@media.mit.edu

• Introduction
  • The Structure from Motion Task
• Applications
• SfM Origins: Photogrammetry and Early Vision
• Linear Approaches
  • Perspective Camera Models
  • Algebraic Projective Geometry
  • Epipolar Geometry
  • The Trifocal Tensor
  • Application to Motion
  • Tomasi-Kanade Factorization
• Issues for Motion Sequences
  • Probabilistic Feature Tracking
  • Causality and Recursion
  • Small Baselines
• Nonlinear Approaches
  • Batch Techniques
  • Filtered and Recursive Multi-Frame Techniques
• The Proposed Nonlinear Recursive Framework
  • The Dynamical System
  • Representation
    • Internal Geometry and Projection Model
    • 3D Structure Model
    • 3D Motion Model - Translation
    • 3D Motion Model - Rotation
    • The Issue of Scale
  • The EKF Implementation
  • Biased Measurements
• Experiments
  • Experiment 1: Egomotion, Models from Video
  • Experiment 2: Object Tracking - Head
  • Independent Evaluation
• Extending the Framework: Feedback and Feedforward Tracking
  • Feeding Back SfM to Low Level Vision
  • Feeding Forward SfM to High Level Vision
• Practical Experience in Commercial Post-Production Applications
• Concluding Remarks
• Acknowledgments
• Bibliography
• About this document ...