Design of a Real-time Crowdsourced Mobility Sensor for Public

Transportation Networks

Marc Pous, Daniel Villatoro, F. Xavier Mercadal and Arol Vi

nolas

Barcelona Digital Technology Center, Barcelona, Spain

Keywords:

Smartcities, Public Transportation Network, Crowdsensing.

Abstract:

Recently, and with the ever-increasing interest in Smart-city paradigm, urban mobility optimization is becom-

ing an active research area. We are specially focused on the real-time optimization of the Public Transportation

Networks, and the passenger ﬂows. In order to achieve such task, it becomes essential to know the state of

the network and its load of passengers in real-time. To the best of our knowledge, no existing infrastructure

is able to keep track of the amount of public transportation users (and their position) in real time. In this

paper we propose a framework architecture that transforms the passengers in active sensors and proﬁt from

the information they provide to estimate the actual usage of the network. In order to create an engaging ex-

perience for passengers (and incentivize the activation of the mobility sensor) we have implemented a higher

level layer which offers users with a trivia-like game, where users compete against other users in real-time,

while providing their position on the network.

1 INTRODUCTION

Catalyzed by the Industrial Revolution, cities have be-

come the acting scenario for the economic tradings

and exchanges that have derived into the modern eco-

nomic system. Due to several factors (such as a reduc-

tion of commuting times and increase the interaction

capabilities), the population within the cities suffered

an outstanding growth and becoming such organiza-

tional structure the predominant and preferred for hu-

man interactions.

With the speciﬁc objective of improving urban

mobility, it is necessary to have a complete under-

standing of the mobility patterns of the citizens. How-

ever, and up to the best of our knowledge, there is

no complete dataset that allows us to obtain such pat-

terns. With the penetration of mobile technologies,

citizens have acquired the capability to continuously

share information anytime anywhere. This continu-

ous sensing capability allows us to combine our sen-

sorized object (the citizens) and the sensor its tracks

its behavior (its mobile device)

Our research focuses on the exploitation of the

human sensing capabilities to understand their mo-

bility patterns in an urban environment, and for the

sake of speciﬁcation we target commuters as our spe-

ciﬁc case of study. Specially, we present the design

of a real-time crowdsourced mobility sensor in public

transportation networks whose usage is incentivized

through gamiﬁcation. A trivia-like game serves as

an incentive for public-transportation users to acti-

vate the designed sensor, and provide our platform

with real-time position information of users, obtain-

ing a complete picture of the state of the network with

up-to-date information. The speciﬁc case of study

that we focus on is the city of Barcelona. Nowadays

the Barcelona transportation agency, among others,

does not have methods neither infrastructure to know

the exact number of people travelling in real time in

its transportation network (composed by bus, metro,

tram and train). With the raise of the computing ca-

pabilities in mobile devices and the appearance of

the Big Data paradigm, it seems appropriate to proﬁt

from the multidisciplinar and interrelated sources of

real-time data and process them in an uniﬁed intel-

ligent core, maximizing the information that can be

obtained from that data. One speciﬁc and interesting

advance proposed is the usage of Case-Base Reason-

ing (CBR) to infere the correct location of users from

the GPS position provided by them when travelling in

the Metro. This is a challenge as the GPS position-

ing return poor positions when the application is used

underground, and therefore, the CBR will correct that

position to what it should really correspond after an

adequate training.

The paper is organized as follows: in section 2

we present a brief summary of the problem and a dis-

cussion related on the improvement of public trans-

231

Pous M., Villatoro D., Mercadal F. and Viñolas A..

Design of a Real-time Crowdsourced Mobility Sensor for Public Transportation Networks.

DOI: 10.5220/0004224402310235

In Proceedings of the 2nd International Conference on Sensor Networks (SENSORNETS-2013), pages 231-235

ISBN: 978-989-8565-45-7

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

portation and available public metrics. In section 3

we describe the modular solution proposed to detect

commuters’ real-time location. Section 4 presents the

actual intelligent analysis tasks that are performed to

overcome some of the problems we have faced. In

section 5 we present some related works and their

contributions to the ﬁeld. In section 6 we summarize

our on-going work for analysing the commuters loca-

tions and state of the Barcelona’s public transporta-

tion network in real-time. Finally, in section 7 we

sketch the next steps that will be taken to improve the

performance of the platform

2 THE PROBLEM: THE LACK OF

REAL-TIME PUBLIC DATA

The optimization of urban processes such as the pub-

lic transportation system has been subject of inter-

est of researchers for many years/for decades (Lynch,

1960; Mandl, 1980; Daganzo, 1997). Some public

transportation networks such as “Transport for Lon-

don” or Seoul metro, have information about the users

entrances and exits controlled through RFID cards

or similar systems, used for optimizing the payment

costs using a “pay-as-you-go” approach. However,

and to the best of our knowledge, there is no existing

infrastructure that provides the load of the network

in real-time. In this paper we sketch the structure of

the architecture that would allow public transportation

users to become active sensors and provide this data

while they travel.

This information is of clear importance to any

public transportation system, as it has been proven

by the numerous costly and time-consuming inter-

views performed to passengers asking about their ori-

gins, destinations, preferred routes and other parame-

ters. Moreover, providing real time information about

the state of the network would increase the aware-

ness of the urban ﬂows within cities through public

transportation systems, and pop up the weakness of

the networks and the behaviours of the commuters in

front of the incidences or big events, amongst many

others.

In this paper, we propose a platform of an integrated

solution that allows the real-time tracking of partici-

pants within the transportation network. To achieve

this task, we rely on the sensing capabilities of pas-

sengers (acquired through their mobile devices such

as smartphones) and the continuous provision of in-

formation. However, we face two problems with this

approach: (1) this sharing implies a cost for the in-

dividual user in terms of battery life and bandwidth

that users are not always willing to sacriﬁce, and (2)

we need to critical mass of users to obtain signiﬁ-

cant sample of information. The individual voluntary

provision of information (for the intelligent analysis

and process in our platform) would represent a long-

term beneﬁt for the users, despite the associated ana-

lyzed costs. However, battery life is a valued resource

that users are always reluctant to share for communal

goods, therefore, we need to provide short term pos-

itive incentives to motivate humans to provide with

this data. Moreover, we need to ensure that a critical

mass of users will participate providing information.

To ensure both problems we opt for the usage of gam-

iﬁcation techniques, translating these incentives into

our platform in terms of a competitive game, that pro-

vides users with distraction during their trips, com-

pensating for the resources consumed, and it is attrac-

tive enough to engage a sufﬁcient amount of users.

As an speciﬁc case of study we want to study

the commuting trends of Barcelona, being the com-

muters a representative portion of the users of the pub-

lic transportation networks. Once we obtain real-time

sample data of mobility, we will be able to understand

the commuting trends and urban ﬂows within the city,

and consequently we would be able to analyze the

weak points that can be reached by us to improve the

system. However, this intelligent processing imply

also the creation and usage of a common language

that refers to each of the units of information treated

in this process.

3 FRAMEWORK

ARCHITECTURE

The solution proposed for the cooperative GPS track-

ing in the Barcelona’s public transportation network

is based on the gamiﬁcation concept to incentivize

users. We deﬁned and developed a game users play

while an underlying application gathers the geoposi-

ton (latitude, longitude and altitude) of the gamers

with the aim of being aware of the Barcelona’s public

transportation usage in a city in real-time. The goal

of the application is the cooperative anonymous real-

time monitorization of the commuters, in the pub-

lic transport network. Commuters seems to be ideal

candidates for this research as they have internalized

the commuting route that they transit everyday, and

their cognitive capabilities can be re-focused on play-

ing our game. While commuters play on their mo-

bile device, they will be competing in real-time with

other players, while sending information to our mo-

bility server. This solution would allow us to track

the whole Barcelona’s public transportation system in

real-time creating an incentive for the commuters to

SENSORNETS2013-2ndInternationalConferenceonSensorNetworks

232

Figure 1: Crowdsourced Public Transportation Sensing

Framework Architecture.

use this game.

The complete architecture (shown in Figure 1)

is formed by three essential interacting components,

which are described below.

• Track-Hunter. The mobile application, that runs

on the users device, gathers the GPS position and

sends it to our server. This component creates a

message that contains the speciﬁc GPS position

(obtained with the native GPS libraries of the mo-

bile device), the timestamp for which that position

was obtained, and the user introduced nickname,

that will allow the continuous track of an speciﬁc

user without compromising his privacy.

• Game Layer. Designed as a incentive mech-

anism, the Track-Hunter application is comple-

mented with a trivia-like game, where players

compete in real-time with other players. The

game itself, combined with the real-time compe-

tition factor creates an engaging and satisfying

interface for which users obtain a trade-off for

the sacriﬁced battery-life consumed by the Track-

Hunter, as well as the attraction to open the appli-

cation every time they commute.

• Mobility Server. The client application sends

(using a REST API

) the gathered information to

a server that recollects all the provided informa-

tion in real-time. This server has been designed

to handle concurrent requests on real-time,

implemented with Node.js

These explained components are essential for the

data acquisition process. However, once the informa-

tion is acquired we need to process it in order to obtain

knowledge about it. This knowledge extraction capa-

bilities are integrated in what we have deﬁned as the

Intelligent Layer. This intelligent layer will be able

to provide speciﬁc information and detailed analysis

Due the simplicity and easiness of the REST protocol

over SOAP and WSDL, most of the new web services are

being build using REST.

http://nodejs.org/

of the commuting trends, speciﬁcally obtaining infor-

mation such as the most used routes, transportation

modes, or peak hours.

4 INTELLIGENT LAYER

In this section we will analyze the different problems

we have faced after the design of the general archi-

tecture. These problems deal with the treatment of

a huge amount of raw data obtained in real-time from

the sensors and its transformation into manageable in-

formation in order to extract knowledge.

4.1 Data Model and Ontology: A

Network Approach

Commuters provide us with a continuous feed of loca-

tions and timestamps, which is almost straightforward

that are associated to movements. However, these

movements are constrained by the actual transporta-

tion infrastructure. This restriction is understood as

a positive point in terms of information storage op-

timization. Moreover, the public transportation net-

work can be easily transported to the mathematical

paradigm of networks and proﬁt from all the theory

and operators developed: stations (where passengers

get in and out) are the nodes of the network, and the

connections amongst stations are the edges. As users

are mobile, each node or edge might have a different

weight. This weight would be changing dynamically,

depending on different interests, such as, distance,

time spent for travelling from one node to another, or

the number of people travelling, amongst other mea-

sures.

These ﬁrst decisions are part of an speciﬁc prob-

lem that we have faced in this challenge, which is the

development of a common speciﬁc vocabulary that

ﬁxes the different types of units and measures that the

system will handle. The objective of the public trans-

portation mobility ontology is twofold: (1) character-

ize the units necessary for our designed optimization

processes, and (2) separate the information acquisi-

tion process from the actual calculations, allowing for

more transparent and differentiated processes that do

not interfere.

Moreover, the proposed ontology is a passenger-

centric ontology as opposed to the majority of the an-

alyzed ontologies in the literature which are mainly

transport centric models (Houda et al., 2010). The on-

tology designed for our platform focuses on the real-

time location of the commuters, their journeys and the

different paths (changes in transportation modes or

transportation lines), within the public transportation

DesignofaReal-timeCrowdsourcedMobilitySensorforPublicTransportationNetworks

233

network. From those journeys it is possible to extract

the stop points and edges, and on the other side, it is

possible to infer the real-time position of the sample

of vehicles with the commuters in them but not taking

into account the schedule or fares.

4.2 Dealing with Poor GPS Signals

Given the precision of the GPS signals obtained from

the users devices, it seems difﬁcult that a mobile de-

vice will provide us with the accurate exact position

that is saved in the reference repository, becoming

therefore an obstacle the association of certain GPS

location to an speciﬁc mobility resource. The most

straightforward solution is to construct a simple Eu-

clidean Distance Filter that given a certain GPS po-

sition and a tolerance level (which determines the ra-

dius of acceptability of distance to the mobility re-

source), it will return the speciﬁc station (or edge) as-

sociated. This initial solution seems to work ﬁne for

the overground transports, but obtaining poor perfor-

mance in underground transportation, given the lack

of precision of the GPS signal.

Our initial test have proven that despite obtaining

a signiﬁcant reduced number of GPS positions when

using underground transportation, some positions are

obtained. Moreover, we do not only have to deal with

the reduced amount of positions, but also with prob-

lems in the precision of the data obtained (e.g. when

being in a certain subway station, the GPS system will

return a GPS position that is 50m. away from that po-

sition). In order to solve this problem we propose a

semi-supervised learning approach, using a Case-base

Reasoning (CBR) technique.

Initially, a knowledge base needs to be gathered,

by transiting the whole network, attaching tags with

the real semantic location (e.g. Metro Station: Diag-

onal, Bus Station: 574, or Edge: Diagonal - Verda-

guer) to every GPS position sent. This knowledge

base will allow the training of the intelligent system,

that will later associate any received GPS position to

the most similar case obtained from the knowledge

base.

4.3 Detecting the Transportation Mode

Strictly related with the previous problem, we will

need to identify the type of transportation that the

user is using, to infer its real location in the network.

Given the different rates of information arrival from

the client application depending on the transport used

(low rates in subway and higher rates in overground),

we propose a simple rule-based classiﬁer, that will in-

telligently categorize the transportation mode of a cer-

tain user from a subset of its provided positions. One

rule example generated by this classiﬁer would be:

IF(Samples with GPS Position > 70 %)

THEN Bus

5 CONCLUSIONS AND FUTURE

WORK

Our contribution proposes a gamiﬁcated cooperative

methodology for tracking urban mobility footprints,

contrasting with traditional survey methods, to im-

prove the understanding of commuters behaviour and

public transportation mobility dynamics.

In the designed system, commuters run an appli-

cation with a competitive game while they are anony-

mously sending their real-time location, allowing us

to know the state of the network and the weak points.

On the other side the application would report real-

time incidences to the users and would recommend

alternative routes to arrive to their usual destination

(in case the system have historical data of user’s des-

tination).

At the moment we only focus on capturing the in-

formation from the commuters and we are aware that

we will be gathering only a subset of potential com-

muters to play our game. Given the openness of the

system, and the nature of any crowdsourced applica-

tion, it is hard at the moment for us to evaluate which

is the validity of our sample. This issue remains as an

open problem to be solved in the future.

As one of the main tasks for the future work, the

server will be given with an intelligent multi-modal

transportation recommender engine which will calcu-

late personalized routes for users in order to optimize

his commuting journey depending on the real-time

status of the public transportation network. As an am-

bitious scientiﬁc challenge, it will be interesting to de-

velop a behavioural model of commuters that could be

integrated with the discoveries provided by other re-

searches about the mobility pattern knowledge of hu-

mans, resulting in a predictive system. This informa-

tion would be highly valuable to understand the over-

all behaviour of such complex system, specially when

facing special events that might produce an overload

of the network.

ACKNOWLEDGEMENTS

This work has been completed with de support of

ACC1

O, the Catalan Agency to promote applied re-

search and innovation.

SENSORNETS2013-2ndInternationalConferenceonSensorNetworks

234

REFERENCES

Daganzo, C. F. (1997). Fundamentals of Transportation

and Trafﬁc Operations. Elsevier Science.

Houda, M., Khemaja, M., Oliveira, K., and Abed, M.

(2010). A public transportation ontology to support

user travel planning. In Loucopoulos, P. and Cavarero,

J.-L., editors, Proceedings of the Fourth IEEE Inter-

national Conference on Research Challenges in Infor-

mation Science, RCIS 2010, Nice, France, May 19-21,

2010, pages 127–136. IEEE.

Lynch, K. (1960). The Image of the City. The MIT Press.

Mandl, C. E. (1980). Evaluation and optimization of urban

public transportation networks. European Journal of

Operational Research, 5(6):396–404.

DesignofaReal-timeCrowdsourcedMobilitySensorforPublicTransportationNetworks

235