Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.772905
Title: Hybrid visual-inertial/magnetic 3D pose estimation for tracking poorly-textured/textureless symmetrical objects
Author: Wibowo, Sapto
ISNI:       0000 0004 7960 3591
Awarding Body: University of Sheffield
Current Institution: University of Sheffield
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
The focus of this research is mainly to develop a visual 3D pose estimation that can be used for many purposes including but not limited to autonomous visual inspection support system. The work overcomes the fundamental problem of region-based pose estimation in tracking poorly-textured/textureless symmetrical objects due to non-unique projection shape given numerous different poses. The work improved the existing state-of-the-art region-based pose estimation, known as Pixel-Wise Posterior 3D Pose estimation (PWP3D), by incorporating with inertial/magnetic orientation estimate. For this purpose, an inertial/magnetic orientation estimate expressed as a full optimisation problem is proposed beforehand. The proposed method, referred to NAG-AHRS, aims to deal better with the non-Gaussian noise and the non-linear model. The NAG-AHRS is then analysed by comparing its output to the motion capture system, as well as benchmarked to five state-of-the-art inertial/magnetic orientation estimates. The experiments show NAG-AHRS outperformed other benchmarking algorithms. Furthermore, NAG-AHRS facilitates the integration to visual-only pose estimation and to develop hybrid visual-inertial/magnetic pose estimation. In contrast with common visual-inertial integration method that has been dominated by Kalman filtering framework, the proposed method integrates visual and inertial/magnetic as a single optimisation problem. The selected optimisation method is Nesterov's Accelerated Gradient (NAG) descent, hence the proposed method is referred to as PWP3Di-NAG. The developed PWP3Di-NAG algorithm is then validated by comparing its output to the reference pose provided by Aruco marker and at the same time, it is also benchmarked to the original PWP3D algorithm. The validation demonstrated some significant performances improvements. Moreover, integrating visual-inertial as a single optimisation problem requires to transform inertial/magnetic measurements into the object reference frame. The required transformation induces an initialisation stage to accurately estimate the initial pose of the object. A novel framework for serving this purpose that combines region-based and edge-based pose estimation in a particle filtering framework is also proposed. The validation shows that the proposed framework be able to estimate the pose of an object with low pose estimation errors.
Supervisor: Dodd, Tony Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.772905  DOI: Not available
Share: