Multi-robot relative localisation using computer vision

This thesis focusses on robots with a marsupial relationship that use computer vision as their main sensor. This work is focused on two separate projects. In computer vision, drift is still a major drawback when using vision-based localisation methods. The first project of this thesis combines...

Full description

Saved in:
Bibliographic Details
Main Author: Ulun, Soner
Other Authors: Wang Han
Format: Theses and Dissertations
Language:English
Published: 2019
Subjects:
Online Access:https://hdl.handle.net/10356/83293
http://hdl.handle.net/10220/50088
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
id sg-ntu-dr.10356-83293
record_format dspace
spelling sg-ntu-dr.10356-832932023-07-04T17:20:02Z Multi-robot relative localisation using computer vision Ulun, Soner Wang Han School of Electrical and Electronic Engineering Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics This thesis focusses on robots with a marsupial relationship that use computer vision as their main sensor. This work is focused on two separate projects. In computer vision, drift is still a major drawback when using vision-based localisation methods. The first project of this thesis combines the position information from two Unmanned Ground Vehicles (UGV) to reduce this drift in GPS denied environments. Relative pose measurements between two robots are used in a similar way to loop-closure. The robots start at a nearby location and share a common field of view. The sharing of the information is done periodically. The pose of each robot is calculated using visual odometry. The relative position of the robots is calculated using both monocular and stereo image pairs, for initial estimation and refining respectively. A specific algorithm is proposed to combine the visual odometry and relative pose estimations from one robot to another in a pose graph optimisation scheme to reduce the drift. Experiments have been conducted on both simulations and real-life. The second project introduces two new methods that enabled the autonomous take-off, tracking and precise landing with an Unmanned Aerial Vehicle (UAV) on a UGV using a dual monocular camera setup. These methods used predefined markers to compute the relative position of the UAV from the landing platform that is attached to the UGV. The cameras on the UAV have different focal lengths, which are providing better performance than using a single camera. Different markers and detectors pairs are analysed and compared to achieve a reliable pose estimation. For a UAV, the most critical parts of the flight are take-off and landing, as this is the time most prone to crashes. Two methods are proposed, named as'Landing calibration' and "Marker Connector", that helped the UAV to land precisely at the desired position in a GPS denied environment. Each method has been verified with more than hundred real-world experiments. Doctor of Philosophy 2019-10-07T01:13:01Z 2019-12-06T15:19:23Z 2019-10-07T01:13:01Z 2019-12-06T15:19:23Z 2019 Thesis Ulun, S. (2019). Multi-robot relative localisation using computer vision. Doctoral thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/83293 http://hdl.handle.net/10220/50088 10.32657/10356/83293 en 176 p. application/pdf
institution Nanyang Technological University
building NTU Library
continent Asia
country Singapore
Singapore
content_provider NTU Library
collection DR-NTU
language English
topic Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics
spellingShingle Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics
Ulun, Soner
Multi-robot relative localisation using computer vision
description This thesis focusses on robots with a marsupial relationship that use computer vision as their main sensor. This work is focused on two separate projects. In computer vision, drift is still a major drawback when using vision-based localisation methods. The first project of this thesis combines the position information from two Unmanned Ground Vehicles (UGV) to reduce this drift in GPS denied environments. Relative pose measurements between two robots are used in a similar way to loop-closure. The robots start at a nearby location and share a common field of view. The sharing of the information is done periodically. The pose of each robot is calculated using visual odometry. The relative position of the robots is calculated using both monocular and stereo image pairs, for initial estimation and refining respectively. A specific algorithm is proposed to combine the visual odometry and relative pose estimations from one robot to another in a pose graph optimisation scheme to reduce the drift. Experiments have been conducted on both simulations and real-life. The second project introduces two new methods that enabled the autonomous take-off, tracking and precise landing with an Unmanned Aerial Vehicle (UAV) on a UGV using a dual monocular camera setup. These methods used predefined markers to compute the relative position of the UAV from the landing platform that is attached to the UGV. The cameras on the UAV have different focal lengths, which are providing better performance than using a single camera. Different markers and detectors pairs are analysed and compared to achieve a reliable pose estimation. For a UAV, the most critical parts of the flight are take-off and landing, as this is the time most prone to crashes. Two methods are proposed, named as'Landing calibration' and "Marker Connector", that helped the UAV to land precisely at the desired position in a GPS denied environment. Each method has been verified with more than hundred real-world experiments.
author2 Wang Han
author_facet Wang Han
Ulun, Soner
format Theses and Dissertations
author Ulun, Soner
author_sort Ulun, Soner
title Multi-robot relative localisation using computer vision
title_short Multi-robot relative localisation using computer vision
title_full Multi-robot relative localisation using computer vision
title_fullStr Multi-robot relative localisation using computer vision
title_full_unstemmed Multi-robot relative localisation using computer vision
title_sort multi-robot relative localisation using computer vision
publishDate 2019
url https://hdl.handle.net/10356/83293
http://hdl.handle.net/10220/50088
_version_ 1772826110676959232