Catalogue : Details

Torsten Sattler

Efficient & Effective Image-Based Localization

ISBN:

978-3-8440-2743-3

Series:

Selected Topics in Computer Graphics
Herausgeber: Prof. Dr. Leif Kobbelt
Aachen

Volume:

Keywords:

Image-Based Localization; Large-Scale Localization; Location Recognition; Image Retrieval; 2D-3D Matching; Prioritized Descriptor Matching; Camera Pose Estimation

Type of publication:

Thesis

Language:

English

Pages:

212 pages

Figures:

51 figures

Weight:

318 g

Format:

21 x 14,8 cm

Bindung:

Paperback

Price:

49,80 € / 62,25 SFr

Published:

May 2014

Buy:

Download:

Available PDF-Files for this title:

You need the Adobe Reader, to open the files. Here you get help and information, for the download.

These files are not printable.


	Document		Abstract
	Type		PDF
	Costs		free
	Action		Display of file - 37 kB (38014 Byte)
	Action		Download of file - 37 kB (38014 Byte)


	Document		Document
	Type		PDF
	Costs		37,35 EUR
	Action		Purchase in obligation and display of file - 4,8 MB (5080734 Byte)
	Action		Purchase in obligation and download of file - 4,8 MB (5080734 Byte)


	Document		Table of contents
	Type		PDF
	Costs		free
	Action		Display of file - 100 kB (102849 Byte)
	Action		Download of file - 100 kB (102849 Byte)

User settings for registered users

You can change your address here or download your paid documents again.

User:	Not logged in.
Actions:	Login / Register Forgotten your password?

Recommendation:

You want to recommend this title?

Review copy:

Here you can order a review copy.

Link:

You want to link this page? Click here.

Export citations:

Text
BibTex
RIS

Abstract:

The problem of image-based localization is the problem of accurately determining the position and orientation from which a novel photo was taken relative to a 3D representation of the scene. It is encountered in many interesting applications such as pedestrian or robot navigation, Augmented Reality, or Structure-from-Motion, creating a strong need for algorithms solving the image-based localization problem. In this thesis, we therefore present solutions to this problem that are both effective and efficient, i.e., we propose methods that can localize novel query images taken under a wide range of viewing conditions while requiring only a small amount of processing time.

We assume that the 3D scene representation is obtained by using Structure-from- Motion techniques to reconstruct the environment from a set of photos. As a result, we can associate each 3D point with multiple image descriptors modeling the local appearance of the scene around this point. We can then obtain 2D-3D correspondences between 2D feature points in the query image and 3D scene points in the model by solving a descriptor matching problem. These 2D-3D matches can in turn be used to estimate the camera position of the query image, i.e., the position and orientation from which it was taken. The main difficulty of descriptor matching lies in the sheer size of the problem, since our models contain millions of 3D points while thousands of features are found in our query images. As a major contribution, we show that the resulting descriptor matching problem can still be solved very efficiently using prioritized search. We propose a prioritization scheme that is easy to implement, yet can be expected to perform close to optimal in practice. By combining our prioritization with a novel active search step that is able to discover additional matches, we are able to derive an imagebasedlocalizationapproachthatachievesorsurpassesstate-of-the-arteffectivenesswhile offering the fastest run-times published so far.

Analyzing such direct matching methods, we demonstrate that their major advantage, namely their ability to identify a set of high-quality matches, also prevents their scalability to larger datasets. Consequently, we also consider image retrieval methods for image-based localization since they are inherently more scalable. As a second major contribution, we identify the algorithmic factors preventing image retrieval methods to achieve the same effectiveness as our original system and propose a modification that is able to close the gap in effectiveness without sacrificing scalability.