Tesis doctoral de Elisabet Batlle Subiros
In the recent years 3d reconstruction has become a very important research topic in computer vision community, involving a wide range of applications such quality control in industry, modeling of large scenarios in robot navigation or reconstruction of buildings, statues and cultural heritage among others. All these tasks rely on surface mapping processes, which imply several steps, including image acquisition, 3d registration and view integration, to obtain the final model of the object/scene. An unsolved problem when registering large objects or scenes resides in the propagation of the error during the registration process, leading to inaccuracies in the alignment of the partial views. This thesis presents a discussion of the 3d registration techniques and proposes a new approach to deal with the misalignment problems, reducing the accumulated error by means of a multiview registration strategy. the first part of this work presents an accurate analysis of the most relevant 3d registration techniques, including initial pose estimation, pairwise registration and multiview registration strategies. A new classification has been proposed, based on both the applications and the approach of the methods that have been discussed. This analysis is the basis of the main contribution of this thesis, since it has allowed to identify the strong and weak points of the existing methods in order to propose a novel contribution that improves the performance of all the previous ones. the second and main contribution of this thesis is the proposal of a new 3d multiview registration strategy. The proposed approach detects revisited regions obtaining cycles of views that are used to reduce the inaccuracies that may exist in the final model due to error propagation. The method takes advantage of both global and local information of the registration process, using graph theory techniques in order correlate multiple views and minimize the propagated error by registering the views in an optimal way. the proposed method has been tested using both synthetic and real data, in order to show and study its behavior and demonstrate its reliability. In addition, some of the most relevant 3d multiview registration methods have also been tested under the same conditions. From the obtained results it can be extracted that the techniques based on graphs provide a good solution to the problem of the propagated error, improving considerably the results obtained by metaview approaches and being more efficient than simultaneous minimization in large scenes. Besides, detecting cycles does not only help improving the process in terms of error propagation, but the use of information related to visited scenes improve significantly the global results.
Datos académicos de la tesis doctoral «Large-scale surface registration«
- Título de la tesis: Large-scale surface registration
- Autor: Elisabet Batlle Subiros
- Universidad: Girona
- Fecha de lectura de la tesis: 19/12/2008
Dirección y tribunal
- Director de la tesis
- Joaquim Salvi Mas
- Tribunal
- Presidente del tribunal: josep Amat girbau
- xavier Binefa valls (vocal)
- el mustapha Mouaddib (vocal)
- emanuele Trucco (vocal)
