 |
Zappella Luca |
   |
| Project:
|
| |
| Supervisor:
|
| |
| Host University:
|
| |
| Personal introduction |
top |
Hi!
The path that led me to Vibot is a bit twisted but eventually I got here and the experience was great. I achieved my master degree in computer science in Italy, at the University of Milan at the end of 2003. Then I enjoyed working for three years, first in a company then as a freelance. I applied for the Vibot master because I was interested in vision and even though I was already working in that field I wanted to learn more, besides the experience of living for two years in three different countries, outside your own one, and with people coming from every corner of the world is something that no one can ever teach you. Now that this adventure is finished I can say I am very pleased I took this twisted path!
|
|
| |
| Project Description and Objectives |
top |
| Motion segmentation is an essential process for many computer vision algorithms. During the last decade, a large amount of work has been trying to tackle this challenge, however, performances of most them still fall far behind human perception. In this thesis the motion segmentation problem is studied, analyzing and reviewing the most important and newest techniques. A classification of all these techniques into different categories according to their main principle and features is proposed, and their strengths and weaknesses are pointed out.
One of the most promising technique emerged from the review, the Local Subspace Affinity is studied in its details highlighting and commenting its critical points. Some improvements to this technique are proposed and validated with tests on synthetic and real sequences. Finally further research directions are suggested. |
|
| |
The approach I took for this work can be deduced by steps planned:
1 - Review the state of the art on motion segmentation
2 - Point out algorithms strengths and weaknesses
3 - Suggest new research directions in order to fill possible gaps in the literature
4 - Choose one of the promising techniques emerged from the discussion on the state of the
art and study it in more detail
5 - Suggest possible improvements, if any
6 - Test the original approach in order to discover its potentials and compare it with the
improved version
7 - Point out future directions connected to this work
|
|
| |
| The tests proved that LSA is a good segmentation
algorithm, furthermore the two improvements proposed
made the algorithm even more robust and able to deal automatically with different noise levels and without knowing how many motions are present in the scene.
The figure below on the left offers a compact at-a-glance overview of the papers studied in this work, while the image on the right shows a sequence (features and trajectories) with three different clusters (red features rotate around a vertical axis and translate from left to right, blue features rotate around an horizontal axis, yellow features remain steady) successfully segmented by the improved LSA.
|
 |
|
| |
| I would like to acknowledge my supervisors and all the teachers of the Vibot master for this awesome experience
|
|
| |
|
MSc Thesis Project documents:
|
|
