Multi-Camera Registration for VR

A flexible, feature-based approach

Master Thesis (2018)
Author(s)

Q. Qian (TU Delft - Electrical Engineering, Mathematics and Computer Science)

Contributor(s)

P.S. Cesar – Mentor

Faculty
Electrical Engineering, Mathematics and Computer Science
Copyright
© 2018 Qinzhuan Qian
More Info
expand_more
Publication Year
2018
Language
English
Copyright
© 2018 Qinzhuan Qian
Graduation Date
14-12-2018
Awarding Institution
Delft University of Technology
Project
VRTogether
Programme
Computer Science
Faculty
Electrical Engineering, Mathematics and Computer Science
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

Real-time point cloud capturing and multiple depth camera 3D reconstruction are vital elements that bring real-time representations into a virtual world and provide an im- mersive experience which can be applied to develop VR/AR applications. To make this possible, camera calibration plays an essential role in providing important camera spa- tial information for 3D scene reconstruction. However, there are still many drawbacks left to improve on camera extrinsic parameters calculation in most existing systems: such as the procedure relies too much on extra calibration markers, or specific depth sensors may have complicated procedures that cannot easily be generalized to other depth sensors.To improve on this, we propose a markerless, feature-based pipeline for multiple cam- era re-calibration. This pipeline contains four main stages. It adopts feature descriptor extracting and matching to solve the issue of requiring additional markers, and the point cloud registration accuracy is improved by using point cloud segmentation and part selection. The experiment results obtained in this research show that this pipeline can calibrate four cameras with a single object (such as a chair, lamp) without the need for additional calibration markers. The extrinsic parameters calculated using this pipeline is more accurate and requires less processing time than originally. This pipeline provides the potential for further human point cloud capturing and camera calibration in real-time 3D reconstruction.

Files

Thesis_Qinzhuan_Qian.pdf
(pdf | 5.57 Mb)
License info not available