Content uploaded by Akira Uchiyama
Author content
All content in this area was uploaded by Akira Uchiyama on Oct 02, 2018
Content may be subject to copyright.
Edge Computing and IoT Based Research for
Building Safe Smart Cities Resistant to Disasters
Teruo Higashino, Hirozumi Yamaguchi,
Akihito Hiromori and Akira Uchiyama
Graduate School of Information Science and Technology
Osaka University, Suita-shi, Osaka 565-0871, Japan
{higashino, h-yamagu, hiromori, uchiyama}@ist.osaka-u.ac.jp
Keiichi Yasumoto
Graduate School of Information Science
Nara Institute of Science and Technology
Takayama, Ikoma 630-0192, Japan
yasumoto@is.naist.jp
Abstract— Recently, several researches concerning with
smart and connected communities have been studied. Soon
the 4G / 5G technology becomes popular, and cellular base
stations will be located densely in the urban space. They
may offer intelligent services for autonomous driving,
urban environment improvement, disaster mitigation,
elderly/disabled people support and so on. Such
infrastructure might function as edge servers for disaster
support base. In this paper, we enumerate several research
issues to be developed in the ICDCS community in the
next decade in order for building safe, smart cities
resistant to disasters. In particular, we focus on (A) up-to-
date urban crowd mobility prediction and (B) resilient
disaster information gathering mechanisms based on the
edge computing paradigm. We investigate recent related
works and projects, and introduce our on-going research
work and insight for disaster mitigation.
Keywords—edge computing; IoT; disaster relief; crowd
mobility prediction
I. INTRODUCTION
Recently, there often occur meteorological disasters caused
by extreme weather such as heavy rain/snowfall, landslide and
forest fires. In earthquake-prone countries such as Japan, East
Asian countries and Mexico, heavy disasters often occur. In
2011, the Great East Japan earthquake caused the massive
tsunami. Building safe and emotionally trusty smart cities
resistant to disasters is a common challenge in the world, and
several national research projects have been launched. NSF
has funded smart and connected communities research [1]. EU
started smart cities & community projects in Horizon 2020.
Japan has advocated super smart society projects called
Society5.0. China has started several smart city related
projects. When large disasters occur, we often suffer severe
damages such as power failure, collapse of buildings, roads
and bridges, fire, transport paralysis, occurrence of
injured/sick, and stranded commuters. It is desirable for local
governments in disaster areas to recognize the situation of
their inhabitants and infrastructure in the early stage so that
safety information and disaster relief information can
smoothly and quickly be delivered to those people who need it.
Along with the progress of the Internet and IoT technology in
recent years, it has become possible to collect and exchange
disaster related information at the time of disaster by using
sensors, smartphones and crowdsourcing technologies.
Several technologies have been developed for speedy and
reliable information grasping and transmission by utilizing
DTN technology, drone technology, image processing
technology utilizing AI and big data analysis.
Although various ICT-based disaster relief technologies
have been developed, there will always be such areas to which
relief information cannot be delivered smoothly. Additionally,
it may take long time to grasp the disaster situation when a
disaster actually occurs. Thus, it is not straightforwardly
performed to accurately grasp the situation of the entire
regions in real time. Furthermore, if a massive blackout occurs
in large cities with millions of people such as New York and
Tokyo, it causes huge damages to transportation networks and
communication infrastructure, which makes grasping the
situation of the afflicted area more difficult. Millions of people
might not be able to return home and there is a possibility that
certain social panic occurs. Many urban infrastructures with
wireless communication functions located in urban districts
(e.g., Chicago’s Array of Things (AoT) [2]) may be used for
autonomous driving, urban environment improvement,
elderly/disabled people support, disaster support and so on.
Such ICT infrastructures (or a part of them) function as edge
servers, and by collaboration of many edge servers, they will
function as a disaster support base for the entire city.
Considering the case that some of them become unavailable
due to blackouts or some other extraordinary situations, it is
necessary to develop an information sensing and aggregation
mechanism based on a new paradigm and an architecture that
allow rescue teams and victims to exchange disaster related
information with high reliability and efficiency. Moreover,
they should be operated over the cloud server and a large
number of edge servers in an autonomous and decentralized
manner. Concretely, we consider that the promotion of the
following research subjects is important:
(A) Up-to-date urban crowd mobility prediction:
Development of technologies to precisely grasp the
crowd mobility including people, vehicles and public
2017 IEEE 37th International Conference on Distributed Computing Systems
1063-6927/17 $31.00 © 2017 IEEE
DOI 10.1109/ICDCS.2017.160
159
2017 IEEE 37th International Conference on Distributed Computing Systems
1063-6927/17 $31.00 © 2017 IEEE
DOI 10.1109/ICDCS.2017.160
1729
transportation in the area of several kilometers in short
time.
(B) Resilient disaster information gathering mechanism:
Development of technologies for (1) sensing and
consolidating information such as road and building
damage at the time of disaster, road blockage and
population distribution quickly and precisely in wide
area, (2) quickly delivering safety information to victims,
and (3) ensuring efficient transportation strategies for
rescue teams and items, and placement strategies of life-
saving emergency workers.
If we can grasp precisely the difference of
pedestrian/vehicle mobility in different time and days such as
weekday/holiday and morning/noon/night, we would realize
the emergency communication infrastructure and sensing
equipment that should be installed beforehand assuming the
situations that cellular networks cannot be used due to power
outages. The research subject (A) is related to urban crowd
mobility prediction. On the other hand, the research subject
(B) is related to speedy and reliable grasp of the damage
situation and information transmission under the assumption
that we can use mobile equipment such as emergency
communication infrastructure, sensing equipment and drones.
The paper is structured as follows. In Section 2, we survey
the existing research work for research subjects (A) and (B).
In Section 3, we enumerate several research subjects to be
developed in the ICDCS community in the next decade.
Section 4 outlines our recent research work related to a
disaster support system. In Section 5, we present the opinions
of the authors on the direction of future research.
II. RELATED WORK
This section surveys the past and ongoing researches and
projects on crowd mobility prediction, disaster information
gathering, and edge computing.
A. Crowd mobility prediction
Floating car data is recently used to estimate the traffic
condition on the road network based on speeds, positions,
directions of driving cars [3]. Floating car data is collected
from driving cars with wireless sensing devices such as
mobile phones. For example, in [4], a method to estimate and
predict the traffic conditions on the highway loop in Rome is
proposed by using floating car data collected from over
600,000 cars every 3 minutes. By applying pattern matching
based on neural networks, it successfully predicts average
speeds after 30 minutes with 3.5-9.5 [km/h] errors. In [5],
floating car data from taxis are used to estimate movement of
taxi passengers. Similar to floating car data, GPS traces of
mobile phone users are exploited to estimate people travels in
urban areas. Specifically, in [6], the authors estimate
movement of evacuees during 1 month after the 2011 Great
East Japan earthquake through analysis of over 9.2 billion
GPS samples. The relationship between people movement and
popular events such as concerts and sport games is also found
by analyzing GPS traces in [7]. Interestingly, in [8], the
authors present a method to estimate user locations by
analyzing twitter texts, focusing on connections between users
in the social network.
Control Signal Records (CSRs) and CDRs (Call Detail
Records) of cellular communications are often used for
mobility analysis and prediction in wide regions. Although
location information of CSRs/CDRs is rather coarse (errors
are usually a few hundreds of meters), it is helpful as they are
collected from a huge number of mobile phone users. The
good feature is that they are automatically generated on the
network operator’s side. In [9], locations of users in near
future are predicted from current phone calls by analyzing the
relationship between occurrences of phone calls, locations and
time.
B. Disaster information gathering
Information delivery is critical during disaster response
[10]. However, disasters can destroy network infrastructure,
leaving the area without end-to-end communication networks.
Because of this, many studies have proposed methods using
Opportunistic and Delay-Tolerant Networks (DTNs) [11].
Pelussi et al. [12] have surveyed opportunistic networks that
are deployed in actual scenarios and discussed different routing
techniques. In [13], the performance during disaster situations
of opportunistic routing protocols is analyzed. DistressNet [14],
[15] is an ad hoc wireless sensor network architecture that
provides data collection services in disaster scenarios. Fujihara
and Miwa proposed an evacuation guidance service using
opportunistic networks [16]. In [17], Kikuchi and Shibata
proposed a disaster information system using mobile cloud
computing. Fajardo et al. proposed an aggregation method to
minimize delays in data delivery [18]. These studies improve
data collection in disaster sites by reducing delivery latency.
For information gathering and provisioning in disaster
situations, the collected information needs to be aggregated and
processed to a meaningful content useful for evacuation and
relief of victims. For this purpose, computation resources are
required even in the situation without the Internet connection.
The concept of opportunistic computing, where users leverage
the available computing resources in the environment has been
presented in [19] and [20]. The Serendipity system [21],[22] is
proposed where nodes distribute computing tasks such as
speech-to-text to remote mobile-devices assuming random
walk mobility models.
C. Edge computing
In a white paper [23], Cisco has predicted that by 2020, 50
billion things will be connected to the Internet that will produce
14.4 trillion dollars in revenue. However, processing data
streams received from such a huge number of devices in a
centralized manner in a cloud server will introduce non-
negligible delays until the service is provided, and this may
reduce service quality and/or rapidly increase service costs due
to wasting cloud computing resources and communication
bandwidth.
1601730
The edge computing [24] and fog computing [25] are based
on the concept of “edge-heavy computing” where data
processing is executed on those components in or on the edge
of networks to mitigate server load. The demerit of these
approaches is the need for investment to replace such network
constituents like Information-Centric networks (ICNs). The
edge centric computing [26], which seeks a more practical
solution by extending edge computing and fog computing,
delegates the processing tasks of cloud servers to other
distributed systems like P2P to realize service components
such as proximity, intelligence, trust and control outside the
cloud.
In [27], we have identified four challenges for treating IoT
data streams in real-time.
C1: Creation technology for IoT data streams: capturing
various real world events anywhere and anytime in a
unified manner.
C2: Networking technology for IoT data streams: enabling
direct flow between producers and consumers in
parallel.
C3: Processing technology for IoT data streams:
processing and aggregating data streams near their
sources.
C4: Content curation technology: intelligently selecting
necessary streams, and processing and integrating
them into valuable contents, based on the interests of
prospective users.
The edge computing paradigm deals mainly with the
challenges C2 and C3, but there is a need for a new paradigm
that appropriately handles all of the above challenges at the
same time.
III. FUTURE RESEARCH PROBLEMS
Below, in order to construct ICT infrastructure for disaster
prevention/mitigation in urban areas of several million people,
we list the research subjects to be developed in the ICDCS
community in the next decade.
A. Crowd mobility prediction related research
As the 4G/5G technology becomes popular, cellular base
stations are closely located in the urban space. They provide
not only WiFi services but also location information provision
for autonomous driving and navigation. Mobile edge
computing based broadband communication services become
possible where many people can use video distribution services
with high bandwidth simultaneously. Autonomous driving is
the most typical distributed system consisiting of many mobile
objects. Related typical research subjects in the ICDCS
community are autonomous driving techniques using deep
learning and big data analysis, real-time (up-to-date)
localization for neighboring vehicles including directly
invisible vehicles, cooperative danger prediction in urban
districts, and mobile edge computing based 3D map
distributions for autonomous driving. For those research
subjects, we should provide efficient and reliable algorithms,
models and theoretical limitation under realistic assumptions.
In Chicago’s Array of Things (AoT) [2], each device is a
communication device combining a processing unit and a
wireless communication unit, and those devices are planned to
be located in urban areas as urban sensing infrastructure. Each
device might be equiped with multiple sensing devices such as
cameras, LIDAR (Laser Imaging Detection and Ranging)
sensors, meteorological/environmental sensors and so on.
Considering at the micro level, it is desirable to continuously
measure and estimate the crowd mobility at stations, malls,
underground shopping malls, and so on with high precision by
using multiple devices such as cameras, LIDAR sensors,
smartphones together. These crowd mobility prediction
information can be used not only for disaster support but also
for various purposes such as improvement of urban
environment, support for elderly/handicapped people,
construction of the next generation commercial/office space
and urban planning. Although it is conceivable to utilize the
trajectory information of smartphone users, various
measurement is required to precisely estimate the mobility of
humans. Also, as shown in Fig. 1, we cannot assume that we
can freely set up many cameras for crowd sensing. Thus, we
need to consider designing and developing crowd mobility
prediction techniques from heterogeneous sensors in urban
districts.
Fig. 1. Crowd Mobility Prediction in Urban Distriction
On the other hand, considering at the macro level, in order
to predict the crowd/vehicle mobility in urban districts, it is
necessary to utilize the floating car data and trajectory
information of smart phones and mobile devices. It is also
conceivable to use surveillance cameras and traffic volume
sensors installed on the roadside. However, storing and
utilizing all floating car data, trajectory information of smart
phones and mobile devices in large cities of several million
people is very costly. In addition, there are cases where it is
difficult to obtain processing data with sufficient accuracy
from companies holding these raw data. In addition, these data
often include privacy information, which causes various social
problems in utilization. It is important to develop (1) new
technology to predict the crowd mobility of the entire target
area using as few sensing equipment as possible, (2)
1611731
technology to complement the crowd mobility of missing areas
from surrounding crowd mobility information, and (3)
technology to estimate the crowd mobility of the entire target
area from spotted crowd mobility information.
Also, suppose that rough mobility information of people
and vehicles of 10 km2 is given as the macro level mobility,
and that detailed mobility information of people and vehicles
of 1 km2 is given as the micro level mobility arranged as 10*10
tiles with various errors and some unknowns. In such a case, it
is important how we guarantee the consistency between the
macro level mobility and micro level one. Especially, the
guarantee of the consistency for OD (origin-destination)
relations between the both levels is an important research issue
when it is difficult to obtain the detailed trajectory information
of people and vehicles for long time.
There are many research challenges in crowd mobility
analysis and prediction. Firstly, there have been two research
trends to obtain crowd density, (i) crowd tracking by
infrastructure and (ii) positioning and context estimation by
individuals’ mobile devices. Although they are considered
independent, unification of such information from different
infrastructure and devices is a significant challenge. As data
from different sources usually have different spatiotemporal
resolutions and scales, they should be mapped onto a global
form of representing crowd densities and mobility. Although
the infrastructure-based tracking using images and videos can
basically obtain crowd distribution in wide regions, only
limited time and location is allowed due to constraints on
privacy-protection, installation locations with view angles. To
complement the time and space not covered by infrastructure,
we should rely on personal devices and cooperative users. The
issue is the accuracy of location/context estimation using
mobile phones. Although they are equipped with various
sensors, they are not sensing-dedicated. This means that daily
tracking should be energy-efficient and opportunistic.
Data reliability is another challenge. While participatory or
opportunistic sensing concept can provide potential solutions,
it needs more investigation on the incentive mechanisms and
distributed strategy planning for complete coverage of the
target region. Confidence of data obtained by anonymous users
is often low, and fake information would often deceive a mass
of people. The trust mechanisms should be introduced to
support secure information delivery even with limited
connectivity to authority.
Finally, mobility prediction should be up-to-date. Since the
information fed by infrastructure and devices is periodically
updated, the mobility prediction should follow up such update.
The data assimilation concept is significantly important
considering the crowd measurement information as input and
update the prediction model. For this purpose, large-scale real-
time simulation is of importance.
B. Disaster information gathering related research
With regard to the collection of disaster information, it does
not mean that it is sufficient to ensure enough bandwidth for
communication networks at the disaster site. Even if a certain
bandwidth can be secured from the disaster site to the base
station in the afflicted area, the transmission efficiency of
disaster information becomes 1/100 if 100 disaster victims
send similar photos and videos for the same fire site to the base
station and/or if they send mails about their own safety to their
families 100 times repeatedly. In a situation where sufficient
bandwidth cannot be secured, a mechanism is required to
aggregate similar disaster information as much as possible and
transmit their aggregated information so that we do not waste
bandwidth. Suppose that we use MapReduce for collecting
disaster information from targeted edge servers. If enough
bandwidth can be ensured, similar photos/videos might be sent
to the master node. However, if enough bandwidth cannot be
ensured, each edge server should aggregate similar
photos/videos and/or return typical one or two to the master
node so that we do not waste crucial bandwidth at disaster
situations. In networks consisting of many edge nodes, the
edge nodes autonomously need to grasp the number of the
disaster victims in the afflicted area and the available
bandwidth for operating an appropriate aggregation mechanism
of disaster information. Research on such mechanisms is also
important. Moreover, if multiple communicable areas are
isolated as a mottled pattern, we need an autonomous
algorithm for efficiently and precisely finding such spots and
creating a mechanism for highly efficient information
transmission among them.
Recently, there are several studies on mechanisms for
relaying/delivering disaster information among such isolated
spots by drones or car-mounted communication equipment. In
a simple method such as assigning one edge server to each spot
that divides a disaster area into multiple spots evenly, if the
amounts of information generated in those spots are
significantly different, the computation loads on some edge
servers might increase. In such a case, the collection/
transmission time of disaster information becomes longer. If
each edge server can grasp the number of victims in the target
area and the damage situation of the infrastructure quickly, an
answer to such load balancing problems is also conceivable.
Researches on such autonomous load distribution and
balancing mechanisms combining simulation technology and
big data technology together are of great importance.
C. Edge computing related research
As we have pointed out in Section II-C, we need a new
paradigm for efficiently processing a massive number of IoT
data streams on edge nodes by solving the following challenges.
C1: Creation technology for IoT data streams
C2: Networking technology for IoT data streams
C3: Processing technology for IoT data streams
C4: Content curation technology
To tackle the challenge C1, for different flows (data
streams), a common metadata format that consists of data type,
granularity, location information, and a set of tags for each
time interval of the flows need to be defined. For the challenge
C2, tags (i.e., contexts, identified objects/events, etc.) need to
be automatically attached to the flows, deriving through a
1621732
learning algorithm. The metadata associated with each flow
facilitates efficient searches and further processing of the flows.
Moreover, for the real-time distribution of flows, a dynamic
granularity adjustment function that reduces the data
granularity is needed.
To tackle the challenge C3, a scalable distributed
computing mechanism among edge nodes is required that
offers functions for executing high-load tasks including
complex event processing and on-line learning among edge
nodes in a distributed and cooperative manner. For distributed
and cooperative processing, predicting processing time for a
heavy task is required. If the predicted time does not satisfy the
time constraint, the task is divided into sub-tasks, which are
sent to nearby edge nodes for execution. Division into sub-
tasks and their allocation must be dynamically done by
considering the available bandwidth in the network and the
computation power of the nearby edge nodes.
To tackle the challenge C4, curation technology must be
developed where multiple flows are compiled to a content
comprehensive to potential users in real-time. To this end, we
need to define a language to describe a curation recipe (e.g., a
task graph to create a content with required spatio-temporal
granularity) along with its execution system. When a curation
recipe is submitted to an edge node, it is executed among
nearby edge nodes in a distributed manner. Therefore, the
execution system is designed and implemented as middleware
with functions such as code/data migration and distributed/
cooperative task processing. Curator aims not only for the
execution of human-edited recipes but also for the support for
recipe-editing work and the further automatic creation of
recipes. For automated curation, it is also needed to realize a
function to measure the value of a content for its prospective
audience and a function to predict new content’s expected
value created with each possible recipe.
Security and privacy issues when treating flows are also
important challenges to be tackled, but we do not address them
in this paper due to space limitation.
IV. OUR ON-GOING RESEARCH WORK
Here, we introduce our on-going research work for disaster
mitigation “Large-Scale, Tempo-Spatial Information Gathering
Mechanism over DTN-enabled Distributed Micro-modules”,
which is supported by Japan Society for the Promotion of
Science (Grant-in-Aid for Scientific Research (S): 26220001).
A. Simple/Portable crowd mobility prediction
It is very necessary in well-populated urban areas to deploy
resistant cellular network infrastructure with uninterruptible
power supply (UPS). However, installation of such
infrastructure is not often reasonable in rural areas. Meanwhile,
recent smartphones and car navigation systems have sufficient
capabilities to communicate and co-operate toward information
gathering and sharing in disaster areas, but these mobile nodes
are not always connected and designing self-organized
computation mechanism is a primary research challenge. This
motivates us to realize information sharing over delay tolerant
networks (DTNs). For the crowd mobility prediction research
(A), we are making the following functions:
(A-1) High precision tracking technology for individuals
and crowds in indoor spaces such as buildings
(A-2) Low infrastructure dependency tracking and
situation awareness technology in indoor, city
districts and public transportation
In our research (A-1), we have developed high precision
tracking technology using LIDAR sensors installed in
commercial buildings and office buildings where highly
accurate crowd sensing and indoor positioning with errors of
several tens of cm order have been realized [28], [29], [30].
Even when a plurality of LIDAR sensors are installed at
multiple locations, we have developed a technology for
practical use that detects moving objects continuously in real
time in consideration of sensor errors. Its prototypes have been
exhibited at the exhibition hall “The Lab.” of a large
commercial building “Grand Front Osaka” in front of Osaka
Station where more than three million people have visited so
far. We also have developed a terminal identification
technology and its system called “Hitonavi” that combines this
system with an acceleration sensor and mobile camera image
of smartphones, and proposed a location based information
sharing method by matching with social media and high
precision tracking.
In our research (A-2), in order to reduce dependency on
infrastructure assuming a disaster situation, we are developing
technology to realize position tracking and peripheral
congestion understanding by utilizing individual smartphone
sensors and mobile wireless communication functions. In
addition to indoor and outdoor environments, we are also
developing practical techniques that can be used to grasp the
situation of public transportation facilities such as stations,
trains, and buses. In [31], we have developed an infrastructure-
free high-speed location identification technology using
multiple smartphones cooperatively. In [32], we have
developed a low power positioning technique based on
cooperative ranging. In [33], cooperative infrastructure-free
trajectory estimation technology has been developed. In [34]
we have proposed a crowd sensing technique using smartphone
camera images where we show that cooperative users’
photographing from a height such as buildings and pedestrian
bridges with a few meters high are effective for estimating the
distribution of crowd on the ground. It shows that crowds are
detected with enough accuracy without any training data. In
[35], we have developed a technique to estimate wide area
mobility for trains and vehicles from LTE communication
history. In [36], we are developing a method to detect
pedestrians’ behavior in public transportation. In [51], we
present a crowd-sensing system for automatic enrichment of
transit stations indoor floorplans. We have also developed train
movement detection technology by wearable device [37],
periphery congestion sensing technology while walking with
smartphones [38], congestion sensing technology while riding
1631733
trains [39], indoor behavior estimation technology and road
surface estimation technology [40], and so on. We also provide
mobile sensing related survey [41]. Now, using these
heterogeneous sensors in urban districts, we are developing a
system to estimate the mobility of people and vehicles in urban
areas in a short time with high accuracy.
B. Disaster information gathering and data analysis
To support disaster information sharing and gathering, we
have been tackling the following three research issues.
(B-1) Sharing and consolidation of sensing data from
vehicles and mobile phones
(B-2) Edge computing technology and optimization of
communication for horizontal data processing and
computing
(B-3) Visualization of cities and human mobility
V2V (Vehicle-to-Vehicle) communication is a basis for
information exchange over DTN. For the first issue (B-1), in
[42], we have presented a scheduling technique for content
exchange over DTN that can be used for efficient, strategic
delivery of information to remote regions.
For the second issue (B-2), most existing IoT platforms do
not fully support distributed, on-site processing. Even for local
services, we need to set up a cloud server and collect/process
data streams in servers, which are far from the data sources.
Such architecture not only limits communication and
computation capacity but also requires additional efforts for
handling privacy-sensitive data, creating barriers to the real-
time utilization of IoT big data. In [27], we have proposed the
Information Flow of Things (IFoT), a new framework for
processing, analyzing, and curating IoT data streams in real
time and a scalable manner based on distributed processing
among IoT devices. In IFoT, both raw data streams and
higher-level streams after processing, aggregating and
merging are called information flows (or flows) and treated
identically. IFoT aims to solve the following three technical
issues: (1) handling various information flows in a unified
manner, (2) processing and analyzing flows in their proximity
and distributing them directly between devices in real time and
a scalable manner, and (3) intelligently integrating different
flows into a content (as a higher-level flow) and providing it in
real time. These issues are solved by three different layered
components: IFoT-Neuron, IFoT-PO3-Engine, and IFoT-
Curator (Fig. 2).
In [43], we have developed IFoT middleware. It provides
four functions: (1) distribution of tasks issued by application
software into subtasks and distributed execution of the sub-
tasks over multiple IoT devices, (2) distribution of data streams
over IoT devices, (3) real-time analysis of the data streams, and
(4) seamless integration of sensors and actuators. Based on this,
in [44], a face recognition system for person tracking was
developed using Raspberry Pi by locally processing video
streams in real time and a distributed manner by using
computational resources of IoT devices. Additionally, toward
automatic curation of multiple data streams, in [45], we
constructed a software agent (machine curator) that compiles
user generated videos that capture a sport game from different
positions/angles/zoom levels automatically to a valuable video
content. For this purpose, we built a machine learning model to
select at each time interval the best one from multiple video
streams that general sports spectators have taken.
Fig. 2. Layered Architecture of IFoT
For the final issue (B-3), we are now developing a
visualization system that incorporates sensing data from
heterogeneous sensing entities such as mobile devices and
vehicles into a single map view. For intuitive recognition of
disaster situation, we have obtained LIDAR point clouds and
images of Osaka University’s entire campus located at Suita
city from an aerial vehicle. To complement the region, which is
hard to see from the sky, we also rely on vehicles with a
LIDAR sensor. Texture mapping is applied onto the 3D model
built from the 3D point clouds. We are now defining the
semantic map information such as streets and passageways that
is mandatory for mobility simulations and prediction.
Moreover, we utilize the edge computing concept together
with DTN technology for disaster response. In [46], we have
developed a computational platform to infer a digital
pedestrian map for disasters from GPS traces collected by
responders exploring the area. In this system, collected data are
sent to edge nodes called computing nodes: commodity
workstations that are deployed in each section of the disaster
area divided according to International Search and Rescue
Advisory Group of the United Nations, for processing. It
establishes a Delay-Tolerant Network (DTN) that uses
Epidemic Routing to communicate across short-ranges and
rescue vehicles as data ferries to communicate across long-
ranges. Moreover, to distribute computation load over multiple
computing nodes, a load balancing heuristic, which uses ferry
route timetables and statistical information about the load of
computing nodes to determine how to offload map inference
tasks. As applications of this platform, we have developed
DTN MapEx [47] that runs on Android terminals to create
disaster maps using DTN communication and MilkCarton [48],
[49] that realizes face image based queries for family tracing
and reuniting when family members evacuated to different
evacuation centers.
C. Smartphone-based crowd and event detection case study
For crowd and event detection, we are now developing a
vision-based people counting technique. The method assumes
1641734
participatory sensing where participants who installed our
smartphone app to obtain images of people from height
locations (such as higher floors of buildings and bridges over
streets) provide photos through the app. We incorporate
lightweight image processing that can be run on the
smartphone, and the app can estimate the number of people in
a photo and reports it to our cloud server. Besides, we provide
a method to analyze tweets and observe correlation between
popular words in the tweets in cyberspace and the presence of
crowd in the physical world to identify the event that might
cause the crowd (see Fig. 3).
Fig. 3. Smartphone-based Crowd and Event Detection Architecture
We have evaluated our system by using images captured
from a live streaming video of Shibuya (Fig. 4), which is one
of the major areas in Tokyo. The video is provided by Shibuya
Television with the permission. We captured images at 2PM
and 8PM on October 30 and 0AM, 1AM and 8PM on October
31. In this evaluation, we focus on the Halloween street party
in Shibuya which was a scheduled event. According to the
information provided by Shibuya Television, we set the height
and the tilted angle of the camera to 9 meters and 18 radians
for our people counting method, respectively. We also
compared the estimated people counts and TF-IDF scores of
the term “Halloween”. Consequently, we see that the estimated
counts and the TF-IDF scores have similar trends. Our system
can successfully detect the Halloween street party event in
Shibuya along with its scale. For more details, readers may
refer to [34].
Fig. 4. Photo from Shibuya Halloween Street Party
D. Realistic human mobility generation case study
We introduce our human mobility generation method using
heterogeneous sensors. The method is used to reproduce
passages, add normal and emergency pedestrian flows and
show their mobility on 2D/3D map. In the proposed method,
we enumerate candidates of pedestrians’ walking routes,
collect the counted numbers of pedestrians at multiple
observation points, and formulate the human mobility
generation problem as an integer linear programming problem.
Using techniques for solving the integer linear programming
(ILP) problem, we mechanically generate the corresponding
pedestrian mobility called Urban Pedestrian Flow (UPF)
mobility [50].
For generating UPF mobility using the method in [50], we
enumerate candidates of pedestrians’ walking routes such as
routeA, routeB and routeC in Fig. 5. Then, we collect the
counted numbers (P1, P2, P3 and P4) of pedestrians at multiple
observation points such as OP1, OP2, OP3 and OP4, respectively.
At the observation point OP1, ideally the counted number P1
should be the same as the sum of the numbers (NA and NC) of
pedestrians of routeA and routeC. Thus, the equation
“NA+NC=P1” should hold. For the four observation points,
similar formulas should hold. Those constraints can be solved
using ILP solvers, and the estimated numbers of all the routes
(NA, NB and NC) are automatically derived. The right figure of
Fig. 5 shows the pedestrian flows in front of JR Osaka Station.
In our experiments, the estimation errors are less than 8% [50].
We have designed and developed a mobile wireless
network simulator called MobiREAL [50]. MobiREAL can
generate UPF mobility, and evaluate the mobility change by
exchanging messages using smartphones and wireless devices.
Fig. 5. Human Mobility Generation
Fig. 6. Reproduction of Emergency Situations
1651735
Using this function, we can reproduce pedestrian mobility,
add normal and emergency pedestrian flows and check
efficiency of evacuation plans. A small fire occurred at
underground Metro Osaka Station in 2012 where a very small
area in a warehouse under the platform is burned due to
electric leak. More than 3,000 people tried to refuge on the
ground from underground. The number of pedestrians varies
depending on weekday or weekend, and rush hour or daytime.
Up-to-date (real-time) crowd control at large stations,
shopping malls and underground malls are very important for
disaster mitigation. Our simulator can be used for checking
efficiency of evacuation plans. Fig. 6 shows visualization of
simulation results for such disaster situations on 3D map [52].
E. Disaster map generation case study
We conducted a realistic simulation-based study when
applying our DTN-based computational platform [46] to a
disaster area of 3 km × 3 km area in Marikina City in the
Philippines, which is often hit by natural disasters. As shown
in Fig. 7, the area is divided into 9 subsections of 1 km × 1 km,
where a rescue team, consisting of 100 disaster responders, is
deployed at each subsection and each responder has a
smartphone with the DTN MapEx application [47]. In each
subsection, we assume that there is a rally point (meeting
place) such as school, medical facility and parish building.
Each rally point possesses an ordinary laptop PC that
functions as a computing node. Rescue vehicles patrol those
rally points serving as message ferries. Here, map inference
denotes the extraction of a graph representing a movable road
network from raw GPS traces of disaster responders based on
the method in [53]. Through distributed computing of map
inference processing at multiple computing nodes and a load
balancing heuristic by the message ferry, we found that our
system can reduce the time to process and deliver subgraphs
of the map to less than 50% in an extreme case where large
quantities of data have to be processed.
Fig. 7. Target Area for Disaster Mapping
V. CONCLUSION
In this paper, we focus on ICT technology for disaster
mitigation and discuss future research subjects in ICDCS
communities. With the rapid progress of the Internet,
smartphones, and IoT devices, the methods for disaster
information gathering/distribution are drastically changing.
Million ordered smartphones and sensors become powerful
means. Several element technologies have been proposed so far.
However, there is still much room for discussion on how to
combine these technologies to construct disaster support
technologies that act effectively in real environments.
ACKNOWLEDGMENT
The research in this paper is partly supported by
MEXT/RIKEN CPS Integrated IT Platform Project and JSPS
Grants-in-Aid for Scientific Research (Grant Numbers:
JP26220001, JP16H01721, JP15H02690, JP16H0291410 and
JP16K12418).
REFERENCES
[1] NSF, “Smart and Connect Communities” Project,
https://www.nsf.gov/pubs/2016/nsf16610/nsf16610.htm
[2] Array of Things, https://arrayofthings.github.io.
[3] T. Toledo and T. Kolechkina, “Estimation of Dynamic Origin –
Destination Matrices using Linear Assignment Matrix Approximations,”
IEEE Trans. on Intelligent Transportation Systems, vol. 14, no. 2, pp.
618-626, 2013.
[4] C. de Fabritiis, R. Ragona and G. Valenti, “Traffic Estimation and
Prediction based on Real Time Floating Car Data,” Proc. of 11th Int.
IEEE Conf. on Intelligent Transportation Systems (ITSC 2008), pp. 197-
203, 2008.
[5] R. Ganti, M. Srivatsa, A. Ranganathan and J. Han, “Inferring Human
Mobility Patterns from Taxicab Location Traces,” Proc. of 2013 ACM
Int. Joint Conf. on Pervasive and Ubiquitous Computing, pp. 459-468,
2013.
[6] T. Horanont, A. Witayangkurn, Y. Sekimoto and R. Shibasaki, “Large-
Scale Auto-GPS Analysis for Discerning Behavior Change During
Crisis,” IEEE Intelligent Systems, vol. 28, no. 4, pp. 26-34, 2013.
[7] F. Calabrese, F. C. Pereira, G. Di Lorenzo, L. Liu and C. Ratti, “The
Geography of Taste: Analyzing Cell-phone Mobility and Social Events,”
Proc. of IEEE Pervasive Computing, pp. 22-37, 2010.
[8] C. A. Davis Jr, G. L. Pappa, D. R. R. de Oliveira and F. de L. Arcanjo,
“Inferring the Location of Twitter Messages based on User
Relationships,” Trans. in GIS, vol. 15, no. 6, pp. 735–751, 2011.
[9] D. Zhang, M. Chen, M. Guizani and H. Xiong, “Mobility Prediction in
Telecom Cloud using Mobile Calls,” IEEE Wireless Communications,
vol. 21, no. 1, pp. 26-32, 2014.
[10] N. Bharosa, J. Lee and M. Janssen, “Challenges and Obstacles in
Sharing and Coordinating Information during Multi-agency Disaster
Response: Propositions from Field Exercises,” Inf. Syst. Front., vol. 12,
no. 1, pp. 49-65, 2010.
[11] K. Fall, “A Delay-tolerant Network Architecture for Challenged
Internets,” Proc. of ACM 2003 Conf. on Applications, Technologies,
Architectures and Protocols for Computer Communications, pp. 27-34,
2003.
[12] L. Pelusi, A. Passarella and M. Conti, “Opportunistic Networking: Data
Forwarding in Disconnected Mobile Ad Hoc Networks,” IEEE
Communication Magagine, vol. 44, no. 11, pp. 134-141, 2006.
[13] A. Martín-Campillo, J. Crowcroft, E. Yoneki and R. Martí, “Evaluating
Opportunistic Networks in Disaster Scenarios”, J. Netw. Comput. Appl.,
vol. 36, no. 2, pp. 870-880, 2013.
[14] S. M. George, W. Zhou, H. Chenji, M. Won, Y.O. Lee, A. Pazarloglou,
R. Stoleru and P. Barooah, “Distressnet: a Wireless Ad Hoc and Sensor
Network Architecture for Situation Management in Disaster Response,”
IEEE Communication Magagine, vol. 48, no. 3, pp. 128-136, 2010.
[15] H. Chenji, W. Zhang, R. Stoleru and C. Arnett, “Distressnet: a Disaster
Response System Providing Constant Availability Cloud-like Services,”
Ad Hoc Networks, vol. 11, no. 8, pp. 2440-2460, 2013.
[16] A. Fujihara and H. Miwa, “Disaster Evacuation Guidance using
Opportunistic Communication: the Potential for Opportunity-based
Service,” Big Data and Internet of Things: A Roadmap for Smart
Environments, Springer, pp. 425-446, 2014.
1661736
[17] Y. Kikuchi and Y. Shibata, “Mobile Cloud Computing for Distributed
Disaster Information System in Challenged Communication
Environment,” Proc. of IEEE 29th Int. Conf. on Advanced Information
Networking and Applications Workshops (WAINA), pp. 512-517, 2015.
[18] J. T. B. Fajardo, K. Yasumoto, N. Shibata, W. Sun and M. Ito, “Disaster
Information Collection with Opportunistic Communication and Message
Aggregation,” Journal of Information Processing, vol. 22, no. 2, pp.
106-117, 2014.
[19] M. Conti, S. Giordano, M. May and A. Passarella, “From Opportunistic
Networks to Opportunistic Computing,” IEEE Communication
Magagine, vol. 48, no. 9, pp. 126-139, 2010.
[20] M. Conti, M. Kumar, et al., “Opportunities in Opportunistic
Computing,” Computer, vol. 43, no. 1, pp. 42-50, 2010.
[21] C. Shi, V. Lakafosis, M. H. Ammar and E. W. Zegura, “Serendipity:
Enabling Remote Computing among Intermittently Connected Mobile
Devices,” Proc. of 13th ACM Int. Symp. on Mobile Ad Hoc Networking
and Computing, pp. 145-154, 2012.
[22] C. Shi, M. H. Ammar, E. W. Zegura and M. Naik, “Computing in Cirrus
Clouds: the Challenge of Intermittent Connectivity,” Proc. of ACM 1st
Edition of the MCC Workshop on Mobile Cloud Computing, pp. 23-28,
2012.
[23] J. Bradley, J. Barbier and D. Handler, “Embracing the Internet of
Everything To Capture Your Share of $14.4 Trillion (White Paper),”
http://www.cisco.com/web/about/ac79/docs/innov/IoE Economy.pdf
(accessed on Nov. 8th, 2015)
[24] A. Davis, J. Parikh and W. E. Weihl, “Edgecomputing: Extending
Enterprise Applications to the Edge of the Internet,” Proc. of 13th Int.
World Wide Web conf. on Alternate Track Papers and Posters, 2004.
[25] F. Bonomi, R. Milito, J. Zhu and S. Addepalli, “Fog Computing and its
Role in the Internet of Things,” Proc. of 1st Edition of the MCC
Workshop on Mobile Cloud Computing (MCC'12), pp. 13-16 (2012).
[26] P. G. Lopez, A. Montresor, D. Epema, A. Datta, T. Higashino, A.
Iamnitchi, M. Barcellos, P. Felber and E. Riviere, “Edge-centric
Computing: Vision and Challenges,” ACM SIGCOMM Computer
Communication Review, vol. 45, no. 5, pp. 37-42, 2015.
[27] K. Yasumoto, H. Yamaguchi, H. Shigeno, “Survey of Real-time
Processing Technologies of IoT Data Streams,” Journal of Information
Processing, vol. 24, no. 2, pp. 195-202, 2016. (invited paper)
[28] T. Higuchi, H. Yamaguchi and T, Higashino, “Trajectory Identification
based on Spatio-temporal Proximity Patterns between Mobile Phones,”
Wireless Networks, vol. 22, no. 2, pp. 563-577, 2016.
[29] T. Higashino and R. Taniguchi, “Recent Research for Social CPS in
Japan,” 2016 US-NSF Cyber-Physical Systems Principal Investigators’
Meeting, 2016.
[30] T. Higuchi, H. Iwahashi, H. Yamaguchi and T. Higashino, “TweetGlue:
Leveraging a Crowd Tracking Infrastructure for Mobile Social
Augmented Reality,” Proc. of 11th Int. Wireless Communications and
Mobile Computing Conf. (IWCMC2015), pp. 1030-1035, 2015.
[31] Y. Noh, H. Yamaguchi and U. Lee, “Infrastructure-free Collaborative
IndoorPositioning Scheme for Time-critical Team Operations,” IEEE
Trans. on Systems, Man and Cybernetics: Systems, 2017. (in press)
[32] T. Higuchi, S. Fujii, H. Yamaguchi and T. Higashino, “Mobile Node
Localization Focusing on Stop-and-Go Behavior of Indoor Pedestrians,”
IEEE Trans. on Mobile Computing, vol.13, no.7, pp. 1564-1578, 2014.
[33] T. Higuchi, H. Yamaguchi and T. Higashino, “Context-Supported Local
Crowd Mapping via Collaborative Sensing with Mobile Phones,”
Pervasive and Mobile Computing, vol. 13, pp. 26-51, 2014.
[34] S. Kojima, A. Uchiyama, M. Shirakawa, A. Hiromori, H. Yamaguchi
and T. Higashino, “Crowd and Event Detection by Fusion of Camera
Images and Micro Blogs,” Proc. of International Workshop on Crowd
Assisted Sensing, Pervasive Systems and Communications (CASPer
2017), pp.1-6, 2017.
[35] Y. Yamada, A. Uchiyama, A. Hiromori, H. Yamaguchi and T.
Higashino, “TravelEstimation using Control Signal Records in Cellular
Networks and Geographical Information,” Proc. of 9th IFIP Wireless
and Mobile Networking Conf. (WMNC 2016), pp. 138-144, 2016.
[36] M. Elhamshary, M. Youssef, A. Uchiyama, H. Yamaguchi and T.
Higashino, “Activity Recognition of Railway Passengers by Fusion of
Low-Power Sensors in Mobile Phones”, Proc. of 23rd Int. Conf. on
Advances in Geographic Information Systems (SIGSPATIAL2015), pp.
1-2, 2015.
[37] T. Higuchi, H. Yamaguchi and T. Higashino, “Tracking Motion Context
of Railway Passengers by Fusion of Low-Power Sensors in Mobile
Devices,” Proc. of 19th ACM Int. Symp. on Wearable Computers
(ISWC2015), pp. 163-170, 2015.
[38] T. Nishimura, T. Higuchi, H. Yamaguchi and T. Higashino, “Detecting
Smoothness of Pedestrian Flows by Participatory Sensing with Mobile
Phones,” Proc. of 2014 ACM Int. Symp. on Wearable Computers (ISWC
2014), pp.15-18, 2014.
[39] Y. Maekawa, A. Uchiyama, H. Yamaguchi and T. Higashino, “Car-level
Congestion and Position Estimation for Railway Trips using Mobile
Phones,” Proc. of 2014 ACM Conf. on Ubiquitous Computing
(UbiComp 2014), pp.939-950, 2014.
[40] T. Satoh, A. Hiromori, H. Yamaguchi and T. Higashino, “A Novel
Estimation Method of Road Condition for Pedestrian Navigation,” Proc.
of Int. Workshop on the Impact of Human Mobility in Pervasive Systems
and Applications (PerMoby2015), pp. 427-432, 2015.
[41] T. Higuchi, H. Yamaguchi and T. Higashino, “Mobile Devices as an
Infrastructure: A Survey of Opportunistic Sensing Technology,” Journal
of Information Processing (JIP), vol. 23, no. 2, pp. 94-104, 2015.
(invited paper)
[42] Y. Adachi, H. Yamaguchi, T. Higashino and T. Umedu, “Cloud-assisted
Dynamic Content Sharing among Vehicles,” Proc. of 6th IEEE Int.
Symp. on Cloud and Service Computing (IEEE SC2-2016), pp. 1-8, 2016.
[43] Y. Nakamura, H. Suwa, Y. Arakawa, H. Yamaguchi, K. Yasumoto,
“Design and Implementation of Middleware for IoT Devices toward
Real-Time Flow Processing,” Proc. of 1st Workshop on Edge
Computing (WEC 2016), 2016.
[44] Y. Nakamura, H. Suwa, Y. Arakawa, H. Yamaguchi, K. Yasumoto,
“Middleware for Proximity Distributed Real-time Processing of IoT
Data Flows,” Proc. of 36th IEEE Int. Conf. on Distributed Computing
Systems (ICDCS 2016), Demo, 2016.
[45] K. Fujisawa, Y. Hirabe, H. Suwa, Y. Arakawa, K. Yasumoto,
“Automatic Live Sport Video Streams Curation System from User
Generated Media,” Int. Journal of Multimedia Data Engineering and
Management (IJMDEM), vol. 7, no. 2, pp. 36-52, 2016.
[46] E. M. Trono, M. Fujimoto, H. Suwa, Y.a Arakawa, K. Yasumoto,
“Generating Pedestrian Maps of Disaster Areas Through Ad-hoc
Deployment of Computing Resources Across a DTN,” Computer
Communications , vol. 100, pp. 129-142, 2017.
[47] E. M. Trono, Y. Arakawa, M. Tamai and K. Yasumoto, “DTN MapEx:
Disaster Area Mapping through Distributed Computing over a Delay
Tolerant Network,” Proc. of 8th Int. Conf. on Mobile Computing and
Ubiquitous Networking (ICMU2015), pp. 179-184, 2015.
[48] E. M. Trono, M. Fujimoto, H. Suwa, Y. Arakawa, K. Yasumoto, “Milk
Carton: Family Tracing and Reunification System using Face
Recognition over a DTN with Deployed Computing Nodes,” Proc. of
2016 Int. Workshop on Information Flow of Things (IFoT 2016), 2016.
[49] M. Trono, M. Fujimoto, H. Suwa, Y. Arakawa, K. Yasumoto, “Milk
Carton: A Face Recognition-based FTR System using Opportunistic
Clustered Computing,” Proc. of 36th IEEE Int. Conf. on Distributed
Computing Systems (ICDCS 2016), Demo, 2016.
[50] K. Maeda, A. Uchiyama, T. Umedu, H. Yamaguchi, K. Yasumoto and T.
Higashino, “Urban Pedestrian Mobility for Mobile Wireless Network
Simulation,” Ad Hoc Networks, vol. 7, no. 1, pp. 153-170, 2009.
[51] M. Elhamshary, M. Youssef, A. Uchiyama, H. Yamaguchi and T.
Higashino, “TransitLabel: A Crowd-Sensing System for Automatic
Labeling of Transit Stations Semantics,” Proc. of 14th ACM Int. Conf.
on Mobile Systems, Applications, and Services (MobiSys 2016), pp.193-
206, 2016.
[52] T. Higashino, “Social CPS Research for Smart and Connected
Communities,” Proc. of 2017 IEEE 5th Int. Electrical Engineering
Congress, pp. K-1 - K-4, 2017. (keynote speech, invited paper)
[53] J. Biagioni and J. Eriksson, “Inferring Road Maps from Global
Positioning System Traces,” Transportation Research Record, vol. 2291,
no.1, pp.61-71, 2012.
1671737