Online Geocoding Services: A benchmarking analysis to some European cities

figures/geofunction-
Introduction
Methods
Implementation
Results
Conclusions
Online Geocoding Services
A benchmarking analysis to some European cities
Gianfranco Di Pietro Fabio Rinnone
Geofunction Srls
Via Luigi Sturzo, 71, 93015 Niscemi CL, Italy
Baltic Geodetic Congress 2017
Gdańsk University of Technology, Gdańsk, Poland
22 - 25 June 2017
Di Pietro G., Rinnone F. Online Geocoding Services

Introduction
Methods
Implementation
Results
Conclusions
Outline
1 Introduction
What is Geocoding?
Hypothesis
2 Methods
Datasets
Benchmark analysis
3 Implementation
GFgeocoder
4 Results
5 Conclusions

Introduction
Methods
Implementation
Results
Conclusions
What is Geocoding?
Hypothesis
Outline
1 Introduction
What is Geocoding?
Hypothesis
2 Methods
Datasets
Benchmark analysis
3 Implementation
GFgeocoder
4 Results
5 Conclusions

Introduction
Methods
Implementation
Results
Conclusions
What is Geocoding?
Hypothesis
What is Geocoding?
Geocoding process consists of translating an address entry and
delevering the best candidate as a point on a map.
Today many Online Geocoding Services offers the possibility to
convert address strings into geographic coordinates (latitude,
longitude).

Introduction
Methods
Implementation
Results
Conclusions
What is Geocoding?
Hypothesis
Hypothesis
We make the following hypothesis:
1 the precise location of house numbers provided by Local
Governments is the most accurate result achievable, because it is
manually performed via a surveying on the ﬁeld executed by an
human operator;
2 it is not possible expecting that the accuracy of geolocation
cannot exceed bounds of 2-3 meters around every point, both for
limits of equipment used for the surveying, and for the real
position of the point, which it can be difﬁcult to be obtained.

Introduction
Methods
Implementation
Results
Conclusions
Datasets
Benchmark analysis
Outline
1 Introduction
What is Geocoding?
Hypothesis
2 Methods
Datasets
Benchmark analysis
3 Implementation
GFgeocoder
4 Results
5 Conclusions

Introduction
Methods
Implementation
Results
Conclusions
Datasets
Benchmark analysis
Datasets
We analyzed the results of geocoding string of addresses published by
following local governments:
Florence (dati.toscana.it): road graph version 1.7.0
available from the open data portal of Tuscan region, Italy.
Cagliari (comune.cagliari.it): database of house
numbers from open data portal of Cagliari, Italy.
Trento (comune.trento.it): database of ﬁrst and second
level house numbers from open data portal of Trento, Italy.
Kristiine, Tallinn (gis.maaamet.ee): over 20k address of
Kriistine district from database of house numbers of Tallin.

Introduction
Methods
Implementation
Results
Conclusions
Datasets
Benchmark analysis
Datasets
From databases we extract tables containing the addresses of strings
and geographic location provided by local governments.
Italian cities uses WGS84 system and Tallinn uses
EPSG:25884-TMBaltic93.
From these tables we randomly extracted address to be passed to the
GFgeocoder (total 30000 strings).

Introduction
Methods
Implementation
Results
Conclusions
Datasets
Benchmark analysis
Datasets
Map of dataset used for the city of Florence.

Introduction
Methods
Implementation
Results
Conclusions
Datasets
Benchmark analysis
Datasets
Map of dataset used for the city of Cagliari.

Introduction
Methods
Implementation
Results
Conclusions
Datasets
Benchmark analysis
Datasets
Map of dataset used for the city of Trento.

Introduction
Methods
Implementation
Results
Conclusions
Datasets
Benchmark analysis
Datasets
Map of dataset used for Kriistine, district of Tallinn.

Introduction
Methods
Implementation
Results
Conclusions
Datasets
Benchmark analysis
Benchmark analysis
We introduce the following parameter.
Definition
Let δ the distance from geocoded position and official dataset position
for each point, we can define:
δ10 =
δi if δi ≤ 10 m (meters)
0 otherwise

Introduction
Methods
Implementation
Results
Conclusions
Datasets
Benchmark analysis
Benchmark analysis
Deﬁnition (Geocoder Approx 10 m)
Ratio (in percentage) of geocoded results in a distance less than 10
meters from position provided by ofﬁcial datasets:
GA10 =
1
n
n
1
Θ(δ10)
GA10 represent a unique value that classify whole addresses investigated. This
parameter is useful for describe the performance of a geocoding service globally for
a location, e.g. a city, a district or a generic buildings dataset.

Introduction
Methods
Implementation
Results
Conclusions
GFgeocoder
Outline
1 Introduction
What is Geocoding?
Hypothesis
2 Methods
Datasets
Benchmark analysis
3 Implementation
GFgeocoder
4 Results
5 Conclusions

Introduction
Methods
Implementation
Results
Conclusions
GFgeocoder
GFgeocoder
GFgeocoder is a multiplatform tool written in Java, executable from
command line and can be run as a background daemon.
It allows loading of input CSV file that contains two columns: first
column, named “address”, must contains addresses and house
numbers, and second column, named “city”, must contains city name.
Output CSV file contains more than two columns: the columns named
“address” and “city” are identical to input file columns; the other
columns contain respectively pairs of coordinates (lat/lon) obtained as
results from Online Geocoders Services.

Introduction
Methods
Implementation
Results
Conclusions
GFgeocoder
GFgeocoder
GFgeocoder uses the following libraries:
Geocoder Java: a set of unofficial Google Java APIs that is
interfaced with official JavaScript Google Maps API v3, via
HTTP requests;
MapQuest: the official set of geocoding libraries offered by
MapQuest. MapQuest offers to developers a RESTful web
service that provides a JSON or XML representation of the
output requested via a HTTP request;
OpenRouteService: an online route planning application based
on open source software that provides a set of public RESTful
APIs that allows user to send geocoding requests.

Introduction
Methods
Implementation
Results
Conclusions
GFgeocoder
GFgeocoder
Class diagram of GFgeocoder.

Introduction
Methods
Implementation
Results
Conclusions
Results
Geocoding services benchmark - GA10
City Google Maps MapQuest ORS Addresses
Trento (IT) 0,04% 0,00% 0,00% 4624
Firenze (IT) 90,05% 0,00% 6,50% 2000
Cagliari (IT) 54,51% 49,12% 21,97% 1311
Kristiine (ES) 0,00% 0,02% 0,00% 22515

Introduction
Methods
Implementation
Results
Conclusions
Results
For Google Maps Geocoder the best performance of GA10 is for Florence addresses,
with 90,05% of string addresses correctly geocoded into 10 meters. The worst is only
Kristiine district of Tallinn with no one address correctly geocoded into 10 meters.
For MapQuest Geocoder the best performance of GA10 is for Cagliari, with 49,12%
of string addresses correctly geocoded into 10 meters. The worst is both Trento city
and Florence with less than 0,005%.
For OpenRouteService the best performance of GA10 is for Cagliari, with 21,7% of
string addresses correctly geocoded into 10 meters. The worst is Kristiine district of
Tallinn with only 4 addresses on 22515 are in a range less than 10 meters.

Introduction
Methods
Implementation
Results
Conclusions
Results
Benchmark results of the addresses of Florence.

Introduction
Methods
Implementation
Results
Conclusions
Results
Benchmark results of the addresses of Trento.

Introduction
Methods
Implementation
Results
Conclusions
Results
Benchmark results of the addresses of Cagliari.

Introduction
Methods
Implementation
Results
Conclusions
Results
Benchmark results of the addresses of Kristiine, Tallinn.

Introduction
Methods
Implementation
Results
Conclusions
Conclusions
The analysis of local spread of geocoding errors produces the
following conclusions:
into old-town areas of Italian cities geocoding errors are least
than the total;
in the particular case of Trento the geocoding errors for suburban
and outskirts areas penalize the overall performance of results;
different performance of geocoding services depends from:
correct interpretation of the addresses strings;
presence of a direct or indirect survey carried out by company
that managed the service.

Introduction
Methods
Implementation
Results
Conclusions
Further work
Further investigations will be conducted to investigate these aspects.
Also, new comparisons will be made using results obtainable via
Nominatim, the open source search and geocoding engine that
consumes OpenStreetMap data.
Finally, the methodology explained in this work is still underway with
data of other international locations.

Online Geocoding Services: A benchmarking analysis to some European cities

Recommended

Recommended

More Related Content

Similar to Online Geocoding Services: A benchmarking analysis to some European cities

Similar to Online Geocoding Services: A benchmarking analysis to some European cities (20)

More from Fabio Rinnone

More from Fabio Rinnone (18)

Recently uploaded

Recently uploaded (20)

Online Geocoding Services: A benchmarking analysis to some European cities