Geometry model for marker-based localisation

Bou Said Yssa, A

Abstract

This work presents a novel approach for position estimation from monocular
vision. It has been shown that vision systems have great capability in reaching precise and accurate measurements and are becoming the state-of-the-art in
navigation. Navigation systems have only been integrated in commercial mobile robots since the early 2000s, and yet localisation in a dynamic environment
that form the main building block of navigation, has no truly elegant solution.
Solutions are many and their strategies and methods differ depending on the
application. For the lack of a single accurate procedure, methods are combined
which make use of different sensors fusion. This thesis focus on the use of monocular vision sensor to develop an accurate Marker-Based positioning system that
can be used in various applications in outdoor, in agriculture for example, and
in other indoor applications. Many contributions arouse here in this context. A
main contribution is in perspective distortion correction in which distortions are
modeled in all its forms with correction process. This is essential when dealing
with measurements and shapes in images. Because of the lack of robustness in
depth sensing using monocular vision-based system, a second contribution is in
the novel spherical marker-based approach position captured, which is designed
and developed within the concept of relative pose estimation. In this Model-Based
position estimation, relative position can be extracted instantaneously without
the need of prior knowledge of the previous state of the camera, as it relies on
monocular image. This model can as well compensate for the lack of knowledge in
the scale of the real world, for example in the case of Monocular Visual Simultaneous Localisation and Mapping (VSLAM). In addition to these contributions, some
experimental and simulation evidence presented in this work has shown feasibility
of the reading measurements like distance capture and relative pose between the
marker-based model and the observer, with reliability and high accuracy. The
system has shown the ability to track accurately the object at a farthest possible
position from low resolution digital images and from a single viewpoint. While
the main application field targeted is tracking mobile-robots, other applications
can profit from this concept like motion capture and application related to the
field of topography.

Citation

Bou Said Yssa, A. Geometry model for marker-based localisation. (Thesis). University Of Salford

• Thesis Type: Thesis
• Deposit Date: Nov 10, 2020
• Publicly Available Date: Nov 10, 2020
• Additional Information: Projects : SMART-E
• Award Date: Oct 30, 2020