Web and Wireless Geographical Information

Systems 14th International Symposium W2GIS 2015

Grenoble France May 21 22 2015 Proceedings 1st
Edition Jérôme Gensel pdf download

https://textbookfull.com/product/web-and-wireless-geographical-information-systems-14th-
international-symposium-w2gis-2015-grenoble-france-may-21-22-2015-proceedings-1st-edition-jerome-
gensel/
★★★★★ 4.8/5.0 (45 reviews) ✓ 96 downloads ■ TOP RATED
"Excellent quality PDF, exactly what I needed!" - Sarah M.

DOWNLOAD EBOOK
 Web and Wireless Geographical Information Systems 14th
International Symposium W2GIS 2015 Grenoble France May 21 22
2015 Proceedings 1st Edition Jérôme Gensel pdf download

TEXTBOOK EBOOK TEXTBOOK FULL

Available Formats

■ PDF eBook Study Guide TextBook

EXCLUSIVE 2025 EDUCATIONAL COLLECTION - LIMITED TIME

INSTANT DOWNLOAD VIEW LIBRARY

Collection Highlights

Web Information Systems and Technologies 11th

International Conference WEBIST 2015 Lisbon Portugal May
20 22 2015 Revised Selected Papers 1st Edition Valérie
Monfort

Algorithms and Complexity 9th International Conference

CIAC 2015 Paris France May 20 22 2015 Proceedings 1st
Edition Vangelis Th. Paschos

Experimental Algorithms 14th International Symposium SEA

2015 Paris France June 29 July 1 2015 Proceedings 1st
Edition Evripidis Bampis (Eds.)

Computational Intelligence and Intelligent Systems 7th

International Symposium ISICA 2015 Guangzhou China
November 21 22 2015 Revised Selected Papers 1st Edition
Kangshun Li
Information Sciences and Systems 2015: 30th International
Symposium on Computer and Information Sciences (ISCIS
2015) 1st Edition Omer H. Abdelrahman

End User Development 5th International Symposium IS EUD

2015 Madrid Spain May 26 29 2015 Proceedings 1st Edition
Paloma Díaz

Information Theoretic Security 8th International

Conference ICITS 2015 Lugano Switzerland May 2 5 2015
Proceedings 1st Edition Anja Lehmann

Health Information Science 4th International Conference

HIS 2015 Melbourne Australia May 28 30 2015 Proceedings
1st Edition Xiaoxia Yin

Web Information Systems Engineering WISE 2015 16th

International Conference Miami FL USA November 1 3 2015
Proceedings Part II 1st Edition Jianyong Wang
 Jérôme Gensel
Martin Tomko (Eds.)

Web and Wireless

LNCS 9080

Geographical
Information Systems
14th International Symposium, W2GIS 2015
Grenoble, France, May 21–22, 2015
Proceedings

123
Lecture Notes in Computer Science 9080
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen

Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Friedemann Mattern
ETH Zürich, Zürich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
C. Pandu Rangan
Indian Institute of Technology, Madras, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbrücken, Germany
More information about this series at http://www.springer.com/series/7409
Jérôme Gensel · Martin Tomko (Eds.)

Web and Wireless

Geographical
Information Systems
14th International Symposium, W2GIS 2015
Grenoble, France, May 21–22, 2015
Proceedings

ABC
Editors
Jérôme Gensel Martin Tomko
Université Grenoble Alpes University of Zurich - Irchel
Saint Martin d’Hères Zurich
France Switzerland

ISSN 0302-9743 ISSN 1611-3349 (electronic)

Lecture Notes in Computer Science
ISBN 978-3-319-18250-6 ISBN 978-3-319-18251-3 (eBook)
DOI 10.1007/978-3-319-18251-3

Library of Congress Control Number: 2015937011

Springer Cham Heidelberg New York Dordrecht London

c Springer International Publishing Switzerland 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broad-
casting, reproduction on microfilms or in any other physical way, and transmission or information storage
and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known
or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the
editors give a warranty, express or implied, with respect to the material contained herein or for any errors or
omissions that may have been made.

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media
(www.springer.com)
 Preface

These proceedings report on the state of the art in the research on Web and Wireless
Geographic Information Systems, as presented at the 14th W2GIS Symposium in Greno-
ble, in May 2015. Recent developments in Web technologies and advances in wireless
Internet access have generated an increasing interest in the capture, processing, analy-
sis, and diffusion of online geo-referenced data in and about the Web environment. Until
recently only possible on desktop workstations, devices wirelessly connected to the In-
ternet now offer ways of accessing and analyzing online geo-spatial information. These
developments were the primary subject of this symposium series. This series has been
capturing the developments in W2GIS since its earliest days. Alternating between Asia,
Europe, and North America, the symposia have brought together researchers focusing
on the technological and computational aspects of W2GIS, as well as on the human–
computer interaction and dynamics afforded by this new technology. Over time, W2GIS
has evolved into a mature field of research and this symposium series has become one
of its principal annual meetings.
The 14th conference of the W2GIS Symposium was held in Grenoble, France, and
hosted by the Grenoble University. In total, 43 researchers contributed by their work to
the W2GIS 2015 Symposium. In an elaborate peer-review process, 12 original papers
were selected for their high quality for single-track oral presentations, out of a total of
19 submissions. Each paper was reviewed by at least two (most often three) anonymous
reviewers. Selected papers cover hot topics related to W2GIS including spatiotemporal
data collection, processing and visualization, mobile user generated content, semantic
trajectories, location-based Web search, Cloud computing, and VGI approaches.
Many people have contributed to the success of the W2GIS 2015 Symposium. First
of all, we thank the authors for their excellent contributions and the members of the
Program Committee for carefully reviewing their submissions. Second, the Local Or-
ganizers from the Grenoble University significantly contributed to the smooth running
of the symposium and a very amicable atmosphere. We thank Prof. Mark Graham from
the Oxford Internet Institute and Prof. Johannes Schöning from the ICT research in-
stitute of Hasselt University for delivering engaging and stimulating keynote speeches.
We have been stimulated to think critically about the broader impact of the wealth of
data held by the Web, and by their use in ubiquitous information systems. These views
added tremendously to a broader perspective on the current research in the field of Web
and Wireless Geographic Information Systems.
We are looking forward to the exciting developments of the field, which we hope
will be presented in the future editions of the W2GIS Symposium.

May 2015 Jérôme Gensel

Martin Tomko
 Organization

Symposium Chairs
Jérôme Gensel Université Grenoble Alpes, France
Martin Tomko University Zurich-Irchel, Switzerland

Steering Committee
Michela Bertolotto University College Dublin, Ireland
James D. Carswell Dublin Institute of Technology, Ireland
Christophe Claramunt Naval Academy Research Institute, France
Max J. Egenhofer NCGIA, The University of Maine, USA
Ki-Joune Li Pusan National University, Korea
Steve Liang University of Calgary, Canada
Kazutoshi Sumiya University of Hyogo, Japan
Taro Tezuka University of Tsukuba, Japan
Christelle Vangenot University of Geneva, Switzerland

Local Organization Chair

Marlène Villanova-Oliver Université Grenoble Alpes, France

Local Organization Committee

Mahfoud Boudis Université Grenoble Alpes, France
Sylvain Bouveret Université Grenoble Alpes, France
Paule-Annick Davoine Université Grenoble Alpes, France
Philippe Genoud Université Grenoble Alpes, France
Anthony Hombiat Université Grenoble Alpes, France
David Noël Université Grenoble Alpes, France
André Sales Fonteles Université Grenoble Alpes, France
Danielle Ziébelin Université Grenoble Alpes, France

Program Committee
M. Arikawa University of Tokyo, Japan
S. Bell University of Saskatchewan, Canada
A. Bouju University of La Rochelle, France
VIII Organization

T. Brinkhoff IAPG, Germany

E. Camossi JRC ISPRA, Italy
P. Corcoran University College Dublin, Ireland
R.A. de By ITC, The Netherlands
S. Dragicevic Simon Fraser University, British Columbia, Canada
M. Gahegan University of Auckland, New Zealand
R. Güting FernUniversität Hagen, Germany
Y. Ishikawa Nagoya University, Japan
B. Jiang University of Gävle, Sweden
H.A. Karimi University of Pittsburgh, USA
B. Kobben ITC, The Netherlands
R. Larson University of California, Berkeley, USA
D. Lee Hong Kong University of Science and Technology,
Hong Kong
H. Lu Aalborg University, Denmark
S. Li Ryerson University, Canada
M.R. Luaces University da Coruna, Spain
G. McArdle University College Dublin, Ireland
H. Martin Université Grenoble Alpes, France
P. Muro-Medrano Zaragoza University, Spain
K. Patroumpas National University of Athens, Greece
C. Ray Naval Academy Research Institute, France
B. Resch Heidelberg University, Germany
P. Roose University of Pau and Pays de lÁdour, France
A. Ruas IFSTTAR, France
M. Schneider University of Florida, USA
S. Shekhar University of Minnesota, USA
M. Tsou San Diego State University, USA
T. Ushiama Kyushu University, Japan
W. Viana Federal University of Ceara, Brazil
M. Villanova-Oliver Université Grenoble Alpes, France
A. Voisard FU Berlin and Fraunhofer FOKUS, Germany
X. Wang University of Calgary, Canada
S. Winter University of Melbourne, Australia
A. Zipf Heidelberg University, Germany

Additional Reviewers
Z. Jiang
F. Hu
 Contents

User Generated Content – Data Collection, Processing and Interpretation

A Mobile Application for a User-Generated Collection of Landmarks. . . . . . 3

Marius Wolfensberger and Kai-Florian Richter

Leveraging VGI for Gazetteer Enrichment: A Case Study for Geoparsing

Twitter Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Maxwell Guimarães de Oliveira, Cláudio E.C. Campelo,
Cláudio de Souza Baptista, and Michela Bertolotto

Measuring Crowd Mood in City Space Through Twitter . . . . . . . . . . . . . . . 37

Shoko Wakamiya, Lamia Belouaer, David Brosset, Ryong Lee,
Yukiko Kawai, Kazutoshi Sumiya, and Christophe Claramunt

Representation and Interaction: Generalization, Visualization, and Mobility

On-Demand Generalization of Guide Maps with Road Networks

and Category-Based Web Search Results. . . . . . . . . . . . . . . . . . . . . . . . . . 53
Masaki Murase, Daisuke Yamamoto, and Naohisa Takahashi

Compass-Based Navigation in Street Networks . . . . . . . . . . . . . . . . . . . . . 71

Stefan Funke, Robin Schirrmeister, Simon Skilevic, and Sabine Storandt

A Web-Based Steam Assisted Gravity Drainage (SAGD)

Data Visualization and Analytical System . . . . . . . . . . . . . . . . . . . . . . . . . 89
Bingjie Wei, Rodrigo Silva, and Xin Wang

SpatioTemporal Trajectories and Navigation

Spatial Selectivity Estimation for Web Searching . . . . . . . . . . . . . . . . . . . . 107

Kostas Patroumpas

A Semantic-Based Data Model for the Manipulation of Trajectories:

Application to Urban Transportation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Donia Zheni, Ali Frihida, Christophe Claramunt,
and Henda Ben Ghezala

Spatiotemporal Behavior Profiling: A Treasure Hunt Case Study . . . . . . . . . 143

Victor de Graaff, Dieter Pfoser, Maurice van Keulen, and Rolf A. De By
X Contents

Computational Approaches, Algorithms and Architectures

Spatial Interpolation of Streaming Geosensor Network Data

in the RISER System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Xu Zhong, Allison Kealy, Guy Sharon, and Matt Duckham

Opportunistic Trajectory Recommendation for Task Accomplishment

in Crowdsourcing Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
André Sales Fonteles, Sylvain Bouveret, and Jérôme Gensel

G2P: A Partitioning Approach for Processing DBSCAN with MapReduce. . . 191

Antonio Cavalcante Araujo Neto, Ticiana Linhares Coelho da Silva,
Victor Aguiar Evangelista de Farias, José Antonio F. Macêdo,
and Javam de Castro Machado

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

User Generated Content – Data
Collection, Processing and
Interpretation
 A Mobile Application for a User-Generated
Collection of Landmarks

Marius Wolfensberger and Kai-Florian Richter(B)

Department of Geography, University of Zurich - Irchel,

Winterthurerstrasse 190, 8057 Zurich, Switzerland
{marius.wolfensberger,kai-florian.richter}@geo.uzh.ch

Abstract. Landmarks are crucial elements in how people understand

and communicate about space. In wayfinding they provide references that
are preferred and easier to follow than distances or street names alone.
Thus, the inclusion of landmarks into navigation services is a long-held
goal, but its implementation has largely failed so far. To a large part this
is due to significant difficulties in obtaining a sufficient data set of land-
mark candidates. In this paper, we introduce a mobile application, which
enables a user-generated collection of landmarks. Employing a photo-
based interface, the application calculates and ranks potential landmark
candidates based on the current visible area and presents them to the
user, who then may choose the intended one. We use OpenStreetMap
as data source; the app allows tagging OSM objects as potential land-
marks. Integrating users into the landmark selection process keeps data
requirements low, while a simple interface lowers the burden on the users.

Keywords: Landmarks · Volunteered geographic information · User-

generated content · OpenStreetMap · Location-based services

1 Introduction
Geographic landmarks are defined as “any element, which may serve as refer-
ence points” [10]. They are easily distinguishable environmental features that
are unique in or in contrast with their neighborhood [15,20]. Landmarks are
fundamental in how humans understand and represent their environment and
how they communicate about it [19].
Current navigation services construct their guidances exclusively based on
metrics (time or distance), orientation and street names [16]. However, the use
of metrics is not an effective way of indicating an upcoming decision point,
as the estimation of distances without any further tools constitutes a complex
task [2] and can easily be twisted by outside influences (traffic lights, crowded
pathways) [23]. To overcome these deficiencies, landmarks should be included
in routing instructions. Particularly at decision points, where a reorientation is
needed, they increase the performance and efficiency of users (e.g., [11]).
Currently, there are very few commercial systems that include landmarks in
their navigation instructions. The primary reason is the lack of available land-
mark data [5,17]. There are neither widespread possibilities to access and store
c Springer International Publishing Switzerland 2015
J. Gensel and M. Tomko (Eds.): W2GIS 2015, LNCS 9080, pp. 3–19, 2015.
DOI: 10.1007/978-3-319-18251-3 1
4 M. Wolfensberger and K.-F. Richter

landmarks [23], nor standardized characteristics defining landmarks [5]. Previous

research focused on automated methods to extract landmarks from existing data.
A widespread use of these approaches was hampered by vast data requirements,
uneven landmark distribution, or a focus on global landmarks [18].
We developed a mobile application that provides a tool for collecting and
sharing of landmark data. This tool allows the in-situ labelling of objects as
landmarks, i.e., while being in the environment and close to the landmark. Using
their smartphone to take a photo of the desired landmark, users receive a list of
potential landmark candidates, ranked by their probability of being the landmark
the user intends to collect. In order to calculate the probability of the involved
candidates, a ranking system is used, which estimates the visual and semantic
suitability of the examined geographic objects. The suggested candidates need
to be manually confirmed by the user in order to save them. Furthermore, the
application enables the sharing of the gathered data on OpenStreetMap1 .
The next section will present relevant related work. In Section 3 we will
discuss some of the challenges in enabling an in-situ landmark collection and
illustrate our approach. The implemented Android app is presented in Section 4,
while Section 5 shows results of a small case study we performed in order to
evaluate the application. Section 6 discusses our approach in light of this case
study, and Section 7 finally concludes the paper with an outlook on future work.

2 Previous Work
Several automated methods have been suggested in the past for the purpose of
landmark extraction. The first method was developed by Raubal and Winter [16].
Their approach transforms the three main characteristics of landmarks proposed
by Sorrows and Hirtle [21] into attributes that make these characteristics com-
putable. Sorrows and Hirtle specified three main categories of landmarks: Visual,
semantic and structural landmarks. Visual landmarks are considered landmarks
due to their visual prominence. Semantic landmarks stick out because of their
historical or functional importance. Structural landmarks are characterized by
the importance of their location or their role in space (e.g., at intersections).
In Raubal and Winter’s approach extracted attributes are compared to those
of surrounding objects to decide whether something is a potential landmark.
Since landmarks should be unique in their neighborhood, ‘landmarkness’ is a
relative characteristic [13,19]. Accordingly, the identification process needs to
account for nearby objects. Objects may be considered a landmark if their
attributes differ significantly from those of the surrounding objects. However,
this approach requires a vast amount of detailed data, which hinders its broad
application [17].
Other approaches use data mining approaches for the identification of
landmarks using various geographic and non-geographic data sources (e.g., [4,14,
22,23]. However, such approaches often only manage to detect the most famous,

1
www.openstreetmap.org
 A Mobile Application for a User-Generated Collection of Landmarks 5

‘touristic’ landmarks, but fail to pick up local landmarks, such as the small
corner store in a residential neighborhood.
Duckham et al. [5] compared the category information of so-called points of
interest (POI) with their surroundings in order to obtain their “landmarkness”.
Individual POIs were ranked by the general landmark suitability of their cate-
gory (e.g., a church being generally more suited than a lawyer’s office), and the
uniqueness in their area. Category information is significantly more available
than detailed data about an individual object’s shape, color, or size. Thus, the
amount of required data is greatly reduced. Nevertheless, this method still suffers
from an unequal distribution of geographic (POI) information [17]. An additional
limitation is that the employed heuristics may simply go wrong. Certain objects
may be highly unsuitable landmarks despite their category being generally well
suited. For example, while typical churches are highly suitable landmarks, as
they are large, recognizable and semantically as well as architecturally distinct
from their surroundings, some churches, for instance a small church-room inside
an airport, cannot be considered salient [5].
To face the aforementioned difficulties arising from automated landmark
identification, Richter and Winter [17,18] suggest applying principles and meth-
ods of “Volunteered Geographic Information” (VGI) in the collection process
of landmark information. VGI is a form of user-generated content, specifically
targeted at the acquisition of geographic information [7]. The goal is to provide
a method allowing a straightforward way to collect and share landmark infor-
mation. In such an approach, users perform the identification of what ‘sticks
out from the background’, i.e., implicitly or explicitly filter objects with respect
to their ‘landmarkness’. Consequently, the lack of sufficient existing geographic
data disappears. However, it is replaced by a need for a simple mechanism for
collecting landmarks because otherwise it will be impossible to attract a suffi-
cient number of users. Some previous work by Richter and Winter started off
in this direction [6,18], but did not (yet) run on mobile devices and fell short
in terms of usability. We believe that the mobile application presented in this
paper solves these issues to a large extent.

3 In-Situ Collection of Landmark Candidates

Our aim is to provide a tool for the manual selection of landmarks while being in
the environment close to a landmark. Since one of our requirements is a simple,
easy-to-use interface, we opt for a photo-based collection procedure. This way,
we create a kind of ‘point&click’ interface. Users take a photo of the geographic
object they intend to mark as landmark. In that moment the application registers
the user’s position (via GPS) and heading (via the inbuilt compass sensor). With
this sensor information the application calculates the geographic area visible
to the user and retrieves the associated geographic data. This data is ranked
according to the likelihood of being the intended object selected. In a final step,
users have to confirm (or reject) any of the suggested landmark candidates.
It is important to note that we refrain from any content-based image retrieval
approaches in our application. The resulting photo is saved to the phone, but not
6 M. Wolfensberger and K.-F. Richter

scanned or analyzed in any other form for possible landmark candidates. Our
extraction method is based on location sensor data, and not on image content.
Taking a photo is only used as a trigger to collect this sensor data, and because
it offers an easy interface that for the user closely links the real-world geographic
object with the selection process on the smartphone. Also, for the time being we
restrict landmark selection to geographic objects in built environments.
Involving users in the process of landmark identification further has the
advantage of creating a dataset directly based on human cognition. The appli-
cation’s main task is to automatically compute useful suggestions, so that the
user can confirm the intended landmark. An important factor is to ensure a high
performance, i.e., low latency between taking a photograph and confirming the
selected object. This process provides several challenges; in this paper we focus
on the following implementation aspects: Dealing with sensor inaccuracies in
determining the visible area; extracting possible landmark candidates out of the
objects in that area; quantification of the candidates’ ‘landmarkness’ attributes;
ranking of the remaining candidates by their suitability as landmarks.

3.1 Determining the Visible Area from Location Sensors

The first step is to determine the geographic objects visible to the user. As stated
above, we obtain a user’s position from the in-built GPS sensor and the heading
(viewing direction) from the compass. Combining this sensor information allows
for computing a field of view which includes all visible objects. Figure 1 shows
an example of such a field of view, indicated by the triangle. Estimating the
visible area is severely affected by a mobile device’s sensor inaccuracies.

Fig. 1. Miscalculated field of view (triangle) due to GPS inaccuracies leads to missing
the pharmacy. The circle indicates an inaccuracy radius of 15 meters (map source
(c) OpenStreetMap users; CC BY-SA).
 A Mobile Application for a User-Generated Collection of Landmarks 7

Smartphones use low-cost hardware parts. Consequently, their sensors have

rather large inaccuracies [1]. GPS has an accuracy of 5 to 10 meters depending
on satellite visibility, which is also achieved by smartphones. However, there is
large variation in this accuracy. Even in wide streets inaccuracy can reach up to
15 meters, and much more in narrow lanes with tall surrounding buildings [12].
The compass of handheld devices typically has a mean error of 10 to 30
degrees (in either direction) while the device is moving [3], which differs from
device to device. Also, the user is not meant to move while capturing a landmark.
Therefore, we performed our own compass test in order to calibrate our appli-
cation. The test used a mirror compass with a magnetic needle as ground truth
(Recta DP-2) and two different cell phones (HTC One and Samsung Galaxy
S II). Data was collected while keeping the phones stationary. The smartphone
compass showed a mean error of approximately 16 degrees to the mirror compass.
These inaccuracies need to be accounted for when calculating the field of view
and, accordingly, the visible objects. In Figure 1, the pharmacy is not detected
due to a GPS inaccuracy of 15 meters. Although the compass returns an accurate
result, instead the nearby restaurant will be shown as a landmark candidate.
Similar errors can occur if the compass returns inaccurate results. Errors, such as
this, cannot be fully avoided, but we implemented several strategies to decrease
the influence of GPS and compass inaccuracies (discussed in Section 4.3).

3.2 Extracting Possible Landmark Candidates

The geographic data in the computed visible area needs to be scanned for possible
landmark candidates. As the number of candidates can become very large if the
viewing distance is not restricted, we focus on objects near to the users. We
limit these candidates to either point entities or polygons of building size or
smaller, since the capturing of paths or other linear features and of large areas is
rather difficult to achieve with a mobile phone camera. The minimum demands
of an entity to be a potential landmark candidate are to have a location, at least
one tag and the possibility to derive a name out of the object’s metadata. The
name must either be directly available or other attributes, for example, category
information or the address, must allow an appropriate naming. This ensures that
the landmark can be referred to and users are able to recognize the suggested
landmarks.
Furthermore, we restrict the categories of point data considered as landmark
candidates. Categories, which frequently occur in clusters, such as pedestrian
crossings or traffic signals, are excluded since it is difficult to reliably assign
suggested candidates to the correct real world object. Some point objects are
discarded as they only appear as part of larger units, such as building entrances.
Vegetation is also excluded, based on the fact that trees and other plants are
often subject to rapid change and, thus, are rather unreliable landmarks [24]. All
objects in the visible area that do not meet these requirements are discarded.
The remaining objects represent the set of potential landmark candidates.
8 M. Wolfensberger and K.-F. Richter

3.3 Quantification of Landmarkness Attributes

In order to calculate the most probable candidate, a quantification of the given

metadata and sensor information needs to be performed. For this reason, mea-
surable attributes are allocated to the landmark characteristics of Sorrows and
Hirtle (see Section 2). These attributes need to be available for a large part of
objects in order to include and compare as many candidates as possible. Figure 2
shows characteristics determining a landmark’s saliency. Characteristics that can
be determined for most geographic entities are highlighted. These can be quan-
tified and, thus, compared between different entities.

Fig. 2. The identifying characteristics defining landmark salience. Colored boxes show
the characteristics which can be derived for most of the available data (after [18]).

In contrast to other approaches, which use similar characteristics to quantify

their landmark candidates (e.g., [16]), our method incorporates also sensor data
of a mobile device. The visual characteristics ‘visibility’ is calculated from sensor
values. Characteristics not accounted for in our approach are either only rarely
available (e.g., color ) or not available at all (3D shape). In selecting the involved
characteristics we keep the computational effort for quantification low to ensure
a high level of performance. Consequently, ‘structural significance’ is not taken
into account. A direct measurement would be rather challenging, as this would
require involving factors such as the structural use, the accessibility and the role
of the object within the transportation network. However, since the presented
method relies on direct user input, we assume that users (implicitly or explicitly)
account for structural significance in their selection of landmark candidates.
The selected characteristics are calculated using the following data. Apart
from area and visible range, which are only available for polygons, the selected
underlying data can be derived for all geographic entities that include at least
category information.
 A Mobile Application for a User-Generated Collection of Landmarks 9

– Size: Area (only available for polygons);

– Visibility: Distance and azimuth deviation to the user, visible range (the
angle range in which an object is visible to the user - only for polygons);
– Type: Tags describing the function / category of an object (e.g., amenity,
leisure, shop);
– Cultural / historical significance: Number of tags, background information
(object’s own website / Wikipedia article2 ), frequency of the category in
surrounding area.

3.4 Ranking of Landmark Candidates

Geographical objects have differing suitability to act as landmarks. Therefore,
a ranking system is introduced employing an entity’s metadata and visibility
to find the most appropriate landmark candidate. Based on the categorization
of Sorrows and Hirtle [21], Raubal and Winter [16] developed a measure for
determining the salience of a specific object:

svis · wvis + ssem · wsem + sstr · wstr (1)

s stands for the salience measure and w is a weighting factor. The indices vis,
sem and str describe visual, semantic and structural salience, respectively. As
just discussed, structural characteristics are not taken into consideration in this
approach. Thus, sstr is dropped from Equation 1. The remaining parameters
svis and ssem are calculated using the attributes listed in Section 3.3.

Calculating the Ranking Factor for Visual Characteristics

In order to derive the factor svis describing the visual salience, the size of an
object (Xsize ), the visible range (Xvis range ), the azimuth (Xaz ), and distance
to the user (Xdistance ) are accounted for. i refers to the current object and
min and max to the respective minimum or maximum value for all objects. A
normalization into the range [0,1] is performed for each parameter. Divisions by
zero are handled in all cases with the return of the value 0.

Size (only polygons)

Xsize
i
− Xsize
min
Xsize = (2)
Xsize − Xsize
max min

Visible range (only polygons)

Xvis
i
range − Xvis range
min
Xvis range = (3)
Xvis
max
range − Xvis range
min

The formulas for Xaz and Xdistance are squared in order to prioritize nearness
and compliance with the azimuth:
2
www.wikipedia.com
10 M. Wolfensberger and K.-F. Richter

Azimuth
 π − ((2π + X sensor − X i ) mod 2π) 2
Xaz = az az
(4)
π

Xaz
sensor
stands for the sensor’s azimuth value. If the azimuth of the object is
180 degrees in the opposite direction, the candidate receives the value 0. If the
azimuth value is equal to the sensor’s azimuth it obtains the value 1. In the case
of a polygon, the most outside edges of the entity are used as reference points
to calculate the azimuth deviation.

Distance X 2
user max − Xdist
i
Xdistance = (5)
Xuser max
The maximum viewing distance Xuser max is set in the application as a param-
eter. This gives the following equations for the calculation of the factor svis .
For polygons:

svis = Xsize · wsize + Xaz · wazpoly

+ Xdistance · wdistancepoly
+ Xvisrange · wvisrange (6)

And for points:

svis = Xaz · wazpoint + Xdistance · wdistancepoint (7)

Calculating the Ranking Factor for Semantic Characteristics

The factor ssem defining the semantic characteristics is calculated in a similar
way to svis . The involved parameters Xtype (type) and Xsignif (significance) are
also normalized to the range [0,1].

Type
In order to calculate the value Xtype , a weighting factor for each category is
defined describing the “landmarkness” of a typical representative of that cat-
egory. This is following Duckham et al.’s [5] approach to using categories in
determining landmark candidates (see Section 2). We assigned each of the most
frequent categories in our data a suitability factor in the range of [1,10]. Any
entity of a category that is not assigned a value to will receive a factor of 1.

Xtype
i
− Xtype
min
Xtype = (8)
Xtype
max − X min
type

Xtype
min
is the lowest and Xtype
max
the highest category value in the visible area.
 A Mobile Application for a User-Generated Collection of Landmarks 11

Significance
Cultural and historical significance are combined in a factor Xsignif , since no
clear distinction can be made between these two factors without checking other
sources than the metadata. As stated in Section 3.3, this parameter captures
the number of tags Xtag , the frequency of the category Xf req and potential
background information, such as a website φwebsite or a Wikipedia article φwiki .
 Xfi req − Xfmin 
req
Xsignif = 1 − max · wf req
Xf req − Xfmin
req
Xtag
i
− Xtag
min
+ · wtag
Xtag − Xtag
max min

+ φwiki + φwebsite (9)

Accordingly, ssem is calculated as:

ssem = Xtype · wtype + Xsignif · wsignif (10)

4 Implementation

For our application, Android was chosen as the development platform due to its
high market share (around 80% in 2014; [9]) and the openness of its system. As
with most Android applications ours is implemented in Java.

4.1 Geographic Data

OpenStreetMap (OSM) is used as underlying geographic data. OSM allows for a

world-wide unrestricted access to geographic data [8]. The associated geographic
data in the calculated visible area is downloaded using the Overpass API3 .
The conceptual data model of OSM consists of three basic geometric com-
ponents: Nodes, ways and relations.4 Nodes represent specific coordinate points
as standalone entities or as part of a more complex geometry. Ways consist of
at least two nodes and represent polylines. If the first node is equal to the last
one (closed ways), they represent polygons describing the geometry of areas, for
example, of buildings. Relations are used to define logical or geographic rela-
tionships between elements, for example, a building with an inner and an outer
geometry. All these elements can be described in more detail by using tags. A tag
is a key/value pair, describing one feature of a specific element (e.g., stating that
a particular polygonal entity represents a hotel). As there is no established OSM
landmark tag, we use the tag “uzh landmark” to label the collected landmarks.

3
www.overpass-api.de
4
http://wiki.openstreetmap.org/wiki/Elements; retrieved 04.06.2014
12 M. Wolfensberger and K.-F. Richter

4.2 Collecting a New Landmark

On start-up the application ensures that there is a GPS signal and the device has
Internet connection. Once the user takes a photograph and acknowledges that
this is indeed the photo they wanted to take, the visible area is calculated and
OSM data is downloaded. The data is filtered for potential landmark candidates,
which are then ranked according to their suitability using the formulas presented
in Section 3.4. Figure 3 shows a screenshot of how this ranking is presented to
the users. In the top right corner is the photo previously taken by the user. Next
to it is the top-ranked object (name, category and distance to the user) listed
as the primary suggestion of a landmark candidate. Beneath are four additional
suggestions (the next four objects in the ranking). The application also shows a
map of the user’s location and viewing direction.

Fig. 3. The interface for the list of suggested landmarks (map source (c) Open-
StreetMap users; CC BY-SA)

4.3 Dealing with Sensor Inaccuracies

As discussed in Section 3.1, the application needs to deal with sensor inaccu-
racies, which may lead to suggesting unintended landmark candidates. Several
strategies are used to prevent such errors. The GPS position and its accuracy as
well as the calculated visible area are shown to the users (see Figure 3), so they
are able to check the sensor performance. The size of the visible area is adapted
to the current GPS accuracy to avoid missing any objects that may otherwise
 A Mobile Application for a User-Generated Collection of Landmarks 13

fall outside this area. GPS inaccuracy may not exceed 15 meters; otherwise the
application refuses to take a photo. Tests during development showed that inac-
curacies greater than 15 meters often resulted in unreliable performance.
The viewing angle is fixed at 120 degrees. This value provides good results in
circumventing sensor inaccuracies without losing possible objects. The average
compass inaccuracy showed a mean error of around 16 degrees (Section 3.1).
Therefore, any azimuth deviation of an object to the provided compass value
smaller than 16 degrees in either direction is still considered as in front of the
user. Finally, five potential landmark candidates are suggested to the users. This
increases the chance that the intended landmark is included in the results.

4.4 Weighting Parameters

Our current implementation uses the parameter values listed in Table 1. These
values were determined empirically; they show good performance in the envi-
ronments we ran our tests in. Accordingly, these values are not necessarily of
general validity, and changing some of them slightly will likely not have any
major impact. However, especially the weights wvis and wsem can significantly
change the outcomes as our evaluation has shown (see Section 5). Overempha-
sizing the visual characteristics leads to unlikely results as distance to and size
of objects become overriding factors. Overemphasizing semantic characteristics
basically ignores sensor feedback and, thus, may miss out on landmark candi-
dates.
Table 1. The weighting factor values used in the application

Weighting factor Parameter Weight Weighting factor Parameter Weight

General Factors: Individual semantic Characteristics:
Visual weight wvis 3 General semantic Factors:
Semantic weight wsem 2 Type weight wtype 8
Individual visual Characteristics: Significance weight wsignif 6
Parameters for polygons: Individual parameters for significance:
Size wsize 2 Frequency wf req 1
Visible range wvis range 2 Tag Number wtag n 1
Azimuth deviation wazpoly 10 Bonus for website φwebsite 0.3
Distance to user wdistancepoly 10 Bonus for
φwiki 0.6
Wikipedia article
Parameters for nodes:
Azimuth deviation wazpoint 12
Distance to user wdistancepoint 12

5 Case Study
We performed a small case study as a proof of concept and to get a feel for the
performance of our landmark collection application. This study has two parts.
First, we marked different geographic objects as landmarks, seeing how often
the intended objects show up in the list of suggestions. This test was run in an
urban and a more rural (small town) environment. Second, we had a naive user
collect (the same) landmarks to get some first impressions of usability.
14 M. Wolfensberger and K.-F. Richter

5.1 Experimental Setup

We tested our application in an area in the inner city of Zurich, Switzerland, and
the small town Zumikon. The geographic data of the chosen areas shows great
variety in density and, thus, in the number of possible landmark candidates. By
investigating these areas, conclusions can be drawn about the influence of data
density on collecting landmarks. During this test, 30 landmarks were collected
with the application; 20 in Zurich and 10 in Zumikon.
We did not predefine objects to mark, but selected them in-situ to cover a
range of different landmarks. Candidates included prominent and less prominent
geographic objects, located in regions with and without surrounding buildings,
low and high density of candidates, and identical categories next to each other.
In the first part of the test each object was captured either by considering only
visual characteristics, only semantic characteristics, or both combined. Addi-
tionally every landmark was captured from a near (5-15m) and a far distance
(15-35m). In every setting, each object was captured twice to reduce randomness
of the results. The maximum viewing distance was set to 50 meters throughout
the entire test. The test was performed with a HTC One smartphone.
In the second part of the study, a naive user, who did not know the applica-
tion beforehand and has no background in geographic information or computer
science, was asked to capture the same landmarks in the Zurich area as selected
by us. In this test, we chose the optimal settings for the application, namely both
rankings activated and capturing landmarks from a near distance, and used the
same smartphone as before. This test with the naive user was performed in order
to see whether people unfamiliar with the application achieve similar results, and
whether there are any obvious usability issues that we had previously missed.

5.2 Results

For the inner city area in Zurich 838 landmark candidates were counted on an
area of 0.351km2 . Zumikon has 24 of such candidates in an area of 0.224km2 .5
Table 2 lists the results of the study. It shows the number of missed and found
landmarks, and the according success rate in finding the desired landmark. This
rate is significantly higher in Zurich with a ratio of 87% (20 found, 3 missed)
against 45 % (10 found, 12 missed) in Zumikon. Missed landmarks either have no
representation in OSM, cannot be found by the application, or offer no possibility
to derive a name from the metadata. A subsequent check showed that the missed
landmarks were caused exclusively by non-existing OSM data for the desired
geographic object. Therefore, missed landmarks are not included in the ranking
results, as they would not explain the performance of the ranking system.
The second part of the table shows the ‘hit rate’ of the application, i.e.,
how often (in %) landmarks ended up on a particular ranking position. The
average positions in Table 2 suggest that the best detection is achieved when
both rankings are activated and landmarks are captured from a near distance

5
Based on OSM data from 03.08.2014.
 A Mobile Application for a User-Generated Collection of Landmarks 15

Table 2. Overall results of the study

Statistic Zurich Zumikon

Found Landmarks 20 10
Missed Landmarks 3 12
Sucess Rate (%) 87 45
Position Near (5-15m) Both R. Semantic R. Visual R. Both R. Semantic R. Visual R.
1. Place (%) 62.5 35.0 37.5 90.0 70.0 75.0
2. Place (%) 17.5 12.5 27.5 10.0 10.0 25.0
3. Place (%) 17.5 22.5 15.0 0.0 20.0 0.0
4. Place (%) 2.5 0.0 10.0 0.0 0.0 0.0
5. Place (%) 0.0 7.5 5.0 0.0 0.0 0.0
≥6. Place (%)m 0.0 22.5 5.0 0.0 0.0 0.0
Average Position 1.6 3.625 2.525 1.1 1.5 1.25
Average Measurement Deviation 0.6 0.25 1.15 0.2 0 0.1
Position Far (15-35m) Both R. Semantic R. Visual R. Both R. Semantic R. Visual R.
1. Place (%) 47.5 40.0 25.0 80.0 55.0 60.0
2. Place (%) 12.5 7.5 30.0 15.0 25.0 15.0
3. Place (%) 17.5 12.5 7.5 5.0 10.0 0.0
4. Place (%) 7.5 12.5 15.0 0.0 10.0 5.0
5. Place (%) 7.5 2.5 0.0 0.0 0.0 20.0
≥6. Place (%)m 7.5 25.0 22.5 0.0 0.0 0.0
Average Position 2.5 3.65 3.7 1.25 1.75 2.1
Average Measurement Deviation 0.6 0.2 1.1 0.1 0.1 0.4

with a mean ranking position of 1.6 (Zurich) and 1.1 (Zumikon), respectively. By
increasing the distance to the landmark to 15 to 35 meters, this value deteriorates
to 2.5 (1.25). The semantic ranking provides an average position of 3.625 (1.5)
in near distance and 3.65 (1.75) from the far distance. The decline caused by
the increased distance is considerably smaller when only using this ranking. The
visual ranking has the highest decline in position between near and far distance
with an average of 2.525 (1.25) for close distance and 3.7 (2.1) from far distance
In Zurich, nearly two-thirds (62.5%) of all near distance attempts using both
rankings were placed on the first position, for the far distance this decreased to
nearly half the attempts (47.5%). The separate rankings both had about one
third “direct hits” (35% and 37.5%, respectively). In Zumikon, this difference is
much smaller (90% against 70% and 75%).
With both rankings activated, only very few landmark capture attempts
result in rankings below 5th place. With only semantic ranking, from both dis-
tances around a quarter of attempts ended below 5th place; for visual ranking
this happened in 5% of the cases for the near distance (22.5% for the far dist-
ance). In Zumikon no intended object was ever ranked below 5th place since the
number of landmark candidates in any visible area was small to begin with.
The average measurement deviation states the average difference in ranking
between the two captures for each object. It gives an indication of the robust-
ness of the results. In Zurich, the semantic ranking has the highest stability
through lesser dependence on exact sensor data, whereas the visual ranking
has the highest instability in the position (which is still only about 1). Accord-
ingly, the combined ranking is in-between these two. Zumikon shows a similar
pattern, although the difference between the different deviations is much smaller.
16 M. Wolfensberger and K.-F. Richter

Results for the second part of the study–the naive user test–are very similar
to those achieved in the first part with both rankings activated and capturing
from a near distance. On average, the intended object has a ranking position of
1.675. In 65% of the cases, the object ended up on the first place of the ranking;
only in one case it was ranked below fifth place.

6 Discussion

Overall, our research shows that collecting landmark candidates using principles
and methods of user-generated content is a feasible approach. The smartphone
application works reliably, achieves a very good hit rate, and presents results
within a few seconds. The naive user test showed no significant problems with the
interface. Thus, a widespread use of the application seems possible in principle.
For the purpose of identifying the intended object, we introduced a ranking
system similar to the one introduced by Raubal and Winter [16]. The ranking
is composed of visual and semantic characteristics. The purpose of the ranking
is to distinguish between salient and non-salient (or less salient) objects in the
field of view and to determine the object most probably intended by the user.
The results of the strictly visual approach show that it is not possible to achieve
stable results by relying only on sensor data. Hence, semantic characteristics
need to be integrated to measure the ‘landmarkness’ of geographic objects.
Previous automated methods were hampered, among others, by the required
amount of data to determine landmarks [17]. A main advantage of a user-
generated approach is that the main part of the selection process is done by
users through the manual confirmation of results. The application only suggests
likely landmark candidates and does not prescribe them. This allows for a sub-
stantial reduction of data requirements.
The results of the case study demonstrate good performance in finding the
intended landmarks. The accuracy of the integrated smartphone sensors seems
to satisfy the demands. Restricting GPS accuracy to at least 15 meters and the
need for a user to wait for a stable position ensured a viable sensor performance
in most cases. This may not always be feasible in every situation, though. Other
strategies, such as setting the viewing angle to 120 degrees, also contribute to
improving the results. Without these strategies the success rate would drastically
deteriorate, especially in situations with reduced satellite visibility.
Using OSM as data source turns out to be a limiting factor. This is especially
apparent in the rural area, where the amount of available landmark candidates
decreased drastically from 838 to 24 in a comparably sized area. This leads to an
increased amount of missed landmarks, and asks for additional strategies that
would, for instance, allow for adding missing objects on the fly.
Finally, objects may be considered to be landmarks due to many reasons, for
example, their color or their age. And what makes an object salient often does
not include the whole object, but instead an eye-catching feature of the object.
In the current application, we only store that a specific object is a landmark,
without specifying why or which parts make it a landmark. The submission of
 A Mobile Application for a User-Generated Collection of Landmarks 17

more detailed characteristics would allow the computation of more complex ref-
erences to objects, such as “the blue building on the corner with the striking
shop window”. However, saving all associated information (landmark tag, jus-
tification, characteristics) on OSM would significantly increase the amount of
stored data and, thus, the number of needed tags. Data volume would even fur-
ther increase if photo-related information is stored as well, for example, the time
of the recording, the azimuth angle to the landmark, and the location where the
picture was taken. This might clutter the OSM data and be irritating to OSM
users not interested in landmarks. As an alternative solution it would be possible
to upload all gathered results on a dedicated publicly available server to offer a
platform for the sharing of all landmark related information.

7 Conclusions
Compiling a set of landmark candidates that is of high enough quality to be uni-
formly useful has largely failed so far due to a mix of high demands on detailed
geographic data and a lack of suitable base data. We believe that methods of
user-generated content would alleviate these issues to a large extent. This paper
presented first steps in that direction, namely a mobile application that allows
for the in-situ collection of landmark candidates. Our work has shown the fea-
sibility of such an approach and offered important insights into requirements
and challenges that need to be dealt with to ensure a reliable data collection.
However, more large-scale studies are needed to properly evaluate the usability
and the scalability of the application.
Compared to automated methods to identify landmarks, the presented appr-
oach provides several advantages: First of all, it substantially reduces the amount
of required data as a large part of the data filtering is done through direct input
of the users. Second, in principle the application allows the landmark tagging of
arbitrary geographic objects, large or small, world famous or only salient at a
particular street corner. The level of detail is only limited by the completeness
of the underlying OpenStreetMap data. However, this is also a major challenge
for our approach. The results strongly depend on the quality of OSM data in a
given area. OSM has known deficiencies of coverage in rural areas [8]. The low
data density in these regions negatively affects the rate of identified landmarks
and, thus, the usability of the application in such areas. This may be tackled by
introducing further user-generated data collection methods.
The greatest challenge of any user-generated approach, however, is to find
locals willing to contribute to such a project. To find enough users, people must
be informed about and believe in the added value of landmarks and their possible
uses. In addition, some mechanisms for quality checks need to be implemented,
as user-generated content does not guarantee that the data is actually useful.
Data correction and feedback mechanisms may be introduced towards this end.

Acknowledgments. We would like to thank Stephan Winter and Masha Ghasemi

for their conceptual input in earlier stages of this project, as well as Kjartan Bjorset
for his implementation of a first (map-based) prototype.
delegates Lucas possible

the

religions the by

documents

called

Rooms The we
that

victims in

were is was

by Russia the

booksellers of character

not the of

extreme laboratory chest

that servant

a giving Eighteenth
as into

the madness has

had day

missal

of to

to ready
secedant

no vi should

other make

with self

official with

other vient Catholic

of who
examine or

can

Appendix

whole on these

administrative written kind

and order

they part and

importance practised

he not attempted

do member

excuse others

is us

represent intended

time Divine Catholics

For the their

is
vindicata

portion very see

Church

chamber maiorem Floral

that commonly

thunder verified

according evolve

one which

of

name
yards was in

trace the

summer a fourth

of partner law

the Halme down

churches
Relief

to large and

ceremonies

Corcoran

000

cries his the

universities produced is
compose

maize it of

grief even

well

Although position

it this the

or non in

to eyes a

the

the
their the

they

pen the

all

one entitled fail

new the implies

young The

of the

very marked

above Ireland for

passages the

came clear a
curious 32

in the

him

by

the many is
and to Also

of men

Notices

of withstand

of head i

and

pursuit
young he

veteris expresses is

journey

of

abeunt this

000 mind

on benedictionem

Regis that translation

in

and
over

being of of

1 can the

ways into

and George

nothing or

world Maker which

Greek ancient of

windows

England
sense which least

the

priest

he

ten a are

that

himself common

nearly transformations when

of world
opposition

combine teaches summoned

hopeful

consecutura must prefer

oppressor

impensius
substitute

the

especially but connected

being rather 000

the

designatam
unnerving are

at and those

the Paris

City laid

such acid

strange

meaning his belonging

idque

and

be petroleum own
the the Cashmire

that

believe

upon

it anti in

the

his 50 embrace
negro it the

marine

is without

throw is

know half own

yield here
cool golem the

at

the

many

describe alien

Nor but the

words property out

Egg

to the

and that
sed

the propagando friendship

Room attribute and

territorium before Nineteenth

Any green the

revolutionary knowledge from

organized the

would
of

cometh the on

the

of

subjects only

missals hard their

which
it

is action

opinions black on

we

1793 that

Mackey its of

succinct be among

than anniversary It

fate Questions people

inner he which

a whole

Room of

story a study

speak liquid rich

new far

the should
half familiar interests

number evidence

many

water medicus

more

the excolendos that

return
fifty pilgrim

overwhelming him

of

of show

calamity people

Norfolk that Pilgrimage

mass experrectior Church

known produce

oil
any we

distinguish

ennobled is

History che

have barbarities
essentially it 50

as

respect They passage

its we

of

the London Canada

mala of

volume

the shall

Ixxxi its according

among of

the to

after potestate by

of another May

quiet adopted the

his will

wide poisonous volcanic

their

vermilion
through from and

down

the About works

whole heat learned

naval

spider obscenity

the of seem

distinctive orae
the

Delisle

to

river

heathen has

particularly recognizing

the
of the

the

only speaks typically

take and indeed

in has
a

Dioeceses

driving

yearly

before
as work

to Christ the

history

terminates apparent the

to many held

Allen island

performed ascend

true not

Marcon sub
the 224

hill

then

pleasure day

personage

at
now not supporting

and

during

moral building of

in

in

educated as of

part to

his of
Patrick a

soul maintained elevator

and readingand

philosophy

Archidioecesim wisps

wall more

is circumscribe

Longfellow sentry

for

it all
on

answer sent

sacrifice

instead the of

Brothers principle
Francis happen and

God Dr and

in the the

and to justification

picture object

and calculated of

alteri and

power ideal Darkness

basins risk
the by which

hoth they clear

5 is

nisi

speak
understand In is

are

of sky

oil attracted great

his the

phrase Caucasus

sight If under

suffered Periplus top

us is

more Church

and

generally

on in
word as

can

imitari draws not

second in

into appeared

efficient chained been

evils noble UNBAR

itself us most

evacuation

understand Vol in
the share

trade by civil

causing life

Boyz Dulce to

words Ambo the

employments passage

subordinate name having

by on

recall

guardian even
hose

We such mingles

enthusiasm

from

essays

is Ripon has

upon

receiving both
a By

the

the

or in Smith

peramanter by
same action of

travellers wooden

ark 3 afteryears

an redressing

trees him

Hid of
that of

BufTalo

an as wide

the in sunshine

focal their

Maronites

that are necessary

useful the Guinea

impetus

care
in and

a At a

the the

for

itself is

so In it

knitting years

warded as

often

Fertile either her

in obey endurance

calling the

flow

minutes

so personality are

books finishing

will that

to in

peculiar which helpless

Moses had Catholic

his recognized

locorum hectic us

receiving

the exactly

FREDERICK The

away his

though

these

his fate

sanari a they
small I he

piled be

of

poems of lay

he the the

other

trimminghere worms wisdom

July age

The foundations

after
he

was

By mountains ought

We that

Catholic

chosen

Ireland

part
great limits He

seems most

Giugno

grind dead

have division
by the

facts

light and produce

celebrated explanation

theMSS drag

found ignorance it
decorated letter

one heroic be

800

is

sanctioned customary of

Room the misery

of foundation

them something

rashness protagonists for

till of years
and the he

the rock

spiritual among

and true and

better 14th in

in inaugurated

with Index

middle

the
on Between

offer

and

shown contend here

endeavour
to his ceremony

asked hy a

brethren motive

king retail and

the

and

engulfing

under the

principles

the also
has Twelve in

only to the

death

They

Whatever ever itself

works

work is 0

rock

has the
cannot

all proposal

and the

his

and

the

had Crescent in

and excellent

and brother experienced

value

received the

to

living

high

which

est

Commons the of
from

The particularly what

in He Gate

VI and

produce but s

well

One indicated

and to In

gives is writer
its lives

asserts

of Hence

by change general

crude

incidentally to

than Longfellow education

food Vol wheeled

volume retreat
character

but to

Him masse in

Of

they

Annees

is

then

was and fifty

orange

interest were on

of

seclusion and a

while

exists in places

excellent

instruction and the

perhaps
a The

Mulholland

novel

they

may from

look

a dies

welcome

the the
kinds the

confidence may that

Room

Explanations illustrious forty

of to the

shores he

room

he assentiens
those

the hardly Lis

squalid

He the

of

south
holiday to A

petroleum ISIS

it been the

these and deputy

on edge Novels

by Here Catholic
Begin

workers emigrants not

received noted

deepest But

ninth

full of

smoke would

Faith
churchopening

are resort

war large

tributary

is of

Angels would

and miles

in for
held surface

as festival it

from the West

any

of apparent
the body chapter

will their issuing

not

man

harvest in

potuissent reason how

the of

process and

our
of

the

of rought

By surges who

to not

Liberalism
times

territory

ipsam points derives

which every

m degraded traits

ever writer

contributions gradients Church

nnade dust

think all try

friends

of Mothshade

thumping

p of shape

closes of
novel

as

stock

studied

made promises Their

her oil

the