Structure from motion

Structure from motion

Structure from motion

Structure from motion

Structure from motion

Structure from motion

Feature tracking

Feature detection

Good features to track

Feature correspondence

Feature distortion

Tracking over many frames

Incorporating dynamics

Modeling uncertainty

Probabilistic Tracking

Kalman filtering:  assume p(x) is a Gaussian

Modeling probabilities with samples

CONDENSATION  [Isard & Blake]

CONDENSATION  [Isard & Blake]

CONDENSATION  [Isard & Blake]

Monte Carlo robot localization

CONDENSATION Contour Tracking

CONDENSATION Contour Tracking

Structure from motion

SFM under orthographic projection

Shape by factorization [Tomasi & Kanade, 92]

Shape by factorization [Tomasi & Kanade, 92]

Singular value decomposition (SVD)

Shape by factorization [Tomasi & Kanade, 92]

Metric constraints

Factorization with noisy data

Many extensions

Extending factorization to perspective

Bundle adjustment

Match Move

Closing the loop

Slide 37

References