Motion Segmentation Results

Images are bigger than shown here. Zoom in or open in a new window for better detail.

Dots are observations (original trajectory data), circles are motion model predictions, residuals are shown as lines from the center of the predictions to the center of the observations. Colored markers are inliers for the corresponding class. Black crosses are trajectories that fit no model (outliers).

Formula 1 Sequences

Unless otherwise noted, these sequences were segmented assuming N=3 (three independent models of rigid motion).

Comments	Results on KLT Tracks
Sequence where two of the three motions have a very limited number of trajectories.
Sequence where the relative motions are very small and where the number of outliers is large. Figure 7.4 and 7.5
Sequence with many outlier trajectories (on the striped line, due to aliasing). Figure 1.1
Sequence with a degenerate motion (the car does not move wrt the camera). Figure 7.4 and 7.5
Same as above but with N=2 independent models of motion. The algorithm chooses to explain the two most dominant motions of the scene and treats trajectories on the helmet as outliers.