Web Page of Tal Nir

PhD student in the computer science department, Technion, Haifa, ISRAEL
Supervisor: Prof. Alfred M. Bruckstein Go to Freddy Bruckstein's web page
 

Robust camera motion estimation

"Causal Camera Motion Estimation by Condensation and Robust Statistics Distance Measures"
PDF version of the paper (Copyright: Springer-Verlag the publisher of ECCV 2004 proceedings)
Authors: Tal Nir (taln@cs.technion.ac.il) and Alfred M. Bruckstein (freddy@cs.technion.ac.il)
Accepted to ECCV 2004

Synthetic movie showing a smooth camera trajectory in a world of 40 stationary points with 20 outliers (random points).
Note that our brain can sense the camera motion as well as to seperate the inliers from the outliers as long as the movie runs.
When the movie stops the seperation of inliers/outliers is lost.
 

The movie sequence analyzed in the paper (Movie 1):

Download Movie 1 in good quality JPEG images

Watch Movie 1

Watch Movie 1 with feature tracking results

Download C code, feature tracking and scene structure data

The features were tracked using a Kanade-Lucas-Tomasi type feature tracker upgraded to color image sequences.

The middle of the letter B is marked by a big yellow cross.
The rest of the features were manually classified:
The inliers are marked as yellow crosses.
The outliers are marked by purple crosses.

Movie 2 (with the same scene structure):

Download Movie 2 in good quality JPEG images

Watch Movie 2

Watch Movie 2 with feature tracking results

Feature tracking data for this movie can be found together with the C code

Movie 3 (with the same scene structure):

Download Movie 3 in good quality JPEG images

Watch Movie 3

Watch Movie 3 with feature tracking results

Feature tracking data for this movie can be found together with the C code

Movie 4 (with the same scene structure):

Download Movie 4 in good quality JPEG images

Watch Movie 4

(No feature tracking data for this movie yet)