Tesis doctoral de Luca Zappella
In this study the problem of motion segmentation is discussed. Motion segmentation aims to decompose a video into the different objects that move throughout the sequence. In many computer vision algorithms this decomposition is the first fundamental step. It is an essential building block for robotics, inspection, video surveillance, video indexing, traffic monitoring and many other applications. The vast amount of literature on motion segmentation testifies to the relevance of the topic. However, the performance of most of the algorithms still falls far behind human perception. in this thesis a review of the main motion segmentation approaches is presented. The main features of motion segmentation algorithms are analysed and a classification of the recent and most important techniques is proposed. specific attention is given to the motion segmentation algorithms that use feature trajectories. these algorithms assume the position of tracked points (features) in each frame of the video sequence as an input. The aim is to group together features that belong to the same motion. The main principle at the base of the proposed algorithms is that trajectories that belong to different motions span different subspaces. Therefore, if the subspace generated by each trajectory could be estimated, and if the different subspaces could be efficiently compared to measure their similarity, the segmentation problem could be cast into a manifold clustering problem. in this study some of the most challenging issues related to trajectory motion segmentation via manifold clustering are tackled. Specifically, new algorithms for the estimation of the rank of the trajectory matrix are proposed. A new measure of similarity between subspaces is presented. Some problems related to the behaviour of principal angles are discussed. Furthermore, a generic tool for the estimation of the number of motions is also developed. The final algorithms that combine the previous proposals are among the very few in motion segmentation literature that do not require any prior knowledge nor manual tuning of their parameters. finally, the last part of the study is dedicated to the development of an algorithm for the correction of an initial motion segmentation solution. Such a correction is achieved by bringing together the motion segmentation and the structure from motion problems. the proposed solution not only assigns the trajectories to the correct motion, but it also estimates the 3d structure of the objects and fills the missing entries in the trajectory matrix. all of the proposed algorithms are tested and compared with state of the art techniques on synthetic and real sequences. Tests show robust behaviour of the proposed algorithms and constant improvements over the state of the art.
Datos académicos de la tesis doctoral «Manifold clustering for motion segmentation«
- Título de la tesis: Manifold clustering for motion segmentation
- Autor: Luca Zappella
- Universidad: Girona
- Fecha de lectura de la tesis: 30/06/2011
Dirección y tribunal
- Director de la tesis
- Xavier Lladó Bardera
- Tribunal
- Presidente del tribunal: emanuele Trucco
- ángel Sappa (vocal)
- fabrice Meriaudeau (vocal)
- jordi Vitri? marca (vocal)