Motion Segmentation Results

The following are image sequences that illustrate the performance of our algorithm. Sequences are linked to specific figures from my thesis Model Fitting for Motion Segmentation [.pdf], where one of the main goals was to introduce an efficient algorithm capable of dealing with sequences with many (over 1,000) noisy trajectories, under the presence of outliers, motion dependencies and motion degeneracies, such as those produced by the widely available KLT feature tracking algorithm. The algorithm was also found to work on datasets with slightly non-rigid motions.

Rigid Motion Sequences
Non-Rigid Motion Sequences