Visual Matching for Robust and Accurate Localization and Mapping

PhD Thesis Proposal Defence


Title: "Visual Matching for Robust and Accurate Localization and Mapping"

by

Mr. Lei ZHOU


Abstract:

An essential component of localization and mapping is visual matching which 
associates 2D observations and 3D entities between individual representations 
of an underlying scene, for example, an image, a point cloud or a reconstructed 
model. The quality of visual matching is critical to the robustness and 
accuracy of localization and mapping. According to the type of data that 
matching algorithms manipulate, we classify the visual matching methodology 
into three categories: 2D-2D matching, 3D-3D matching and 2D-3D matching, which 
will be addressed in the thesis, respectively. The problem of 2D-2D matching 
focuses on the identification of visual overlap and feature correspondences 
between pairwise images. Particularly, we seek to improve 2D-2D matching in the 
case of significant scale changes, and propose a scale-invariant matcher to 
tackle the large scale variation of views. The 3D-3D matching is closely 
related to point cloud registration which requires a set of accurate 
correspondences between points in 3D space. Since the matching results could be 
contaminated by outliers, a robust matching approach based on a graphical model 
is developed for outlier rejection. The 2D-3D matching is typically applied in 
camera relocalization for accurate pose determination. Rather than handling 
one-shot relocalization which can be non-robust in many situations, we propose 
to learn temporally-consistent 2D-3D matching to estimate the pose of each 
video frame in sequence by considering the time dependency explicitly. Finally, 
to reach a localization or mapping result with better accuracy, we propose a 
stochastic bundle adjustment algorithm which refines the geometry globally  at 
scale based on the visual matches.


Date:			Friday, 25 October 2019

Time:                  	12:00noon - 2:00pm

Venue:                  Room 5504
                         lifts 25/26

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


**** ALL are Welcome ****