[4eyes] Fwd: MS Project Defense -- Jack Alexander

[Tobias Hollerer]

Today, 1pm.

Cheers,

     Tobias



-------- Forwarded Message --------
Subject: 	[FACULTY] MS Project Defense -- Jack Alexander
Date: 	Fri, 17 Aug 2018 10:13:14 -0700
From: 	benji <benji at cs.ucsb.edu>
Reply-To: 	benji at cs.ucsb.edu
To: 	faculty at lists.cs.ucsb.edu <faculty at lists.cs.ucsb.edu>, 
grads at lists.cs.ucsb.edu <grads at lists.cs.ucsb.edu>, 
lecturers at lists.cs.ucsb.edu <lecturers at lists.cs.ucsb.edu>, 
research at lists.cs.ucsb.edu <research at lists.cs.ucsb.edu>



MS Project Defense

*Jack Alexander*

Tuesday, August 21st, 2018

1:00pm – HFH 1132

*Committee: *Tobias Hollerer (Chair), Matthew Turk

*Title:* Alignment of Small-Scale 3D Captures for Wide-Area Point Cloud 
Reconstruction

*Abstract:*

We present a set of tools for the construction of large-scale point 
cloud models from smaller local-area captures. The systems developed 
include a backend server infrastructure for generating structure from 
motion (SFM) models; an efficient, user-friendly system for the manual 
alignment of point clouds models; and an automatic post-processor that 
uses the iterative closest point (ICP) algorithm to improve on 
user-provided alignment.

This system focuses on supporting point clouds generated using a variety 
of capture modalities, including those that do not record color data. 
This limits the methods available for automatic registration and 
presents a particular challenge for ICP due to the varying density of 
the input point clouds. Additionally, the noisiness of the point clouds 
we are interested in aligning and their relatively small degree of 
overlap distinguishes this effort from many applications of the 
iterative closest point algorithm. We present an analysis of the range 
of successful initial configurations (SIC-range) of ICP when applied to 
two SFM-generated point clouds and to point clouds generated using 
different modalities. Our novel contributions include techniques for 
applying ICP to partially overlapping point clouds and a demonstration 
that ICP can improve on manual user alignment when the area of overlap 
is small.

Everyone welcome!

-- 
Tobias Hollerer
Professor, Department of Computer Science
University of California, Santa Barbara, CA 93106-5110

holl at cs.ucsb.edu, Office: (805)284-9395, Fax: (805)893-8553

-------------- next part --------------
_______________________________________________
faculty mailing list
faculty at lists.cs.ucsb.edu
https://lists.cs.ucsb.edu/mailman/listinfo/faculty