This thesis is focused on three dimensional data model reconstruction and on the robot movement estimation computed from data captured during its trajectory. Our main objective is the automatic 3D environment map reconstruction using the six degrees of freedom robot movement estimation. Furthermore, our solutions have been developed for application in real time, while the robot is moving. A lot of work has been done in order to obtain a way for reducing input 3D data complexity into a pre-processing stage. This data model has to be as complete as possible.
Its independence from the physical system used is also one of our objectives. For this reason, all the proposals included in this thesis are designed for working with raw 3D data. In this way, we increase the application scope of our solutions since there are no constraints about the physical system used for grabbing data. Moreover, the application boundaries of our methods have been studied by means of a huge
set of experiments, performed into real scenarios. Working conditions have been also
established.
This thesis is divided into two parts. The first one is focused on the construction of
a 3D data model built from the scene that has been captured by the robot. After the
study of several modeling approaches, we propose two solutions. First, we describe a
method for extracting the main planar surfaces of a 3D scene. This model is complete
and accurate, and can be used in some applications such as tele-presence, virtual
reality, architectonic reconstruction, etc. On the other hand, we propose a method
that estimates planar surface pathes into the scene in a quicker way. This method is
not so accurate as the previous one, but can be used in time restricted mobile robotic
applications, for example, to estimate the robot movement at the same time as it
is realized. This first part is completed with a study for adding new features to the
resulting 3D model. In this way, we present how to partially use the process performed
for obtaining planar patches in order to obtain creases into the scene. Furthermore,
plane extraction procedure error has been studied.
The second part of this thesis is focused on robot movement estimation problem.
In order to obtain this movement, we use its 3D environment information. Instead
of using 3D raw data captured by robot sensors, we use the planar based 3D model
computed with our quicker modeling method. Our proposal for robot movement estimation
is a modification of the Iterative Closest Points (ICP) algorithm, but instead
of points, we use planar patches for registration. One of the most important features
of our proposal is its robustness in the presence of outliers. The correct work of our
proposal is demonstrated by means of a huge number of experiments, performed both
in indoors and outdoors, and using different kind of 3D sensors. The results obtained
bring us to the conclusion that our method can be used into dynamic semi-structured environments.
Finally, we present all the conclusions obtained during this thesis and the future
work. The later is focused on obtaining a more complete 3D model that could be used
for estimating robot movements. We also plan to include robot movement estimation
into a global error rectification algorithm. |
