Interactive Techniques for Visual Retrieval and Matching in 3D reconstruction applications

MPhil Thesis Defence


Title: "Interactive Techniques for Visual Retrieval and Matching in 3D 
reconstruction applications"

By

Mr. Ziyu WANG


Abstract

3D reconstruction applications have been constantly developed in the past 
decades in aspects including accuracy, efficiency and robustness. In the 
reconstruction pipeline, structure from motion (SfM) is a photogrammetric 
approach to calculate 3D point positions and camera poses based on 2D images. 
Nowadays, the output results of SfM are still sensitive to the input images. 
Unqualified images, such as those under diverse illumination conditions or with 
insufficient details, may lead to inadequate feature point matches and 
therefore cause deficiencies in the reconstruction output. In order to further 
improve the quality of the result, we propose interactive techniques, in which 
humans are able to supplement data required by SfM pipeline to refine the 
reconstruction.

Our proposed method consists of two main parts: a graphical user interface 
(GUI) designed for match visualization, which assists users to localize weak 
image connections and create supplementary matches; an improved SfM pipeline 
with interfaces which can incorporate manual matches and retrieve new images. 
With these techniques, the reconstruction result can be improved interactively.


Date:			Friday, 17 January 2020

Time:			2:00pm - 4:00pm

Venue:			Room 5501
 			Lifts 25/26

Committee Members:	Prof. Long Quan (Supervisor)
 			Prof. Chiew-Lan Tai (Chairperson)
 			Dr. Pedro Sander


**** ALL are Welcome ****