Content uploaded by Afaf Mosaif
Author content
All content in this area was uploaded by Afaf Mosaif on Apr 22, 2021
Content may be subject to copyright.
A New System for Real-time Video Surveillance in Smart
Cities Based on Wireless Visual Sensor Networks and Fog
Computing
Afaf Mosaif and Said Rakrak
Laboratory of Computer and Systems Engineering (L2IS), Faculty of Science and Techniques, Cadi Ayyad University,
40000 Marrakesh, Morocco
Email: afaf.mosaif@edu.uca.ac.ma; s.rakrak@uca.ma
Abstract—Nowadays, public security is becoming an
increasingly serious issue in our society and its requirements
have been extended from urban centers to all remote areas.
Therefore, surveillance and security cameras are being deployed
worldwide. Wireless Visual Sensor Networks nodes can be
employed as camera nodes to monitor in the city without the
need for any cables installation. However, these cameras are
constrained in processing, memory, and energy resources. Also,
they generate a massive amount of data that must be analyzed in
real-time to ensure public safety and deal with emergency
situations. As a result, data processing, information fusion, and
decision making have to be executed on-site (near to the data
collection location). Besides, surveillance cameras are
directional sensors, which makes the coverage problem another
issue to deal with. Therefore, we present a new system for real-
time video surveillance in a smart city, in which transportations
equipped with camera nodes are used as the mobile part of the
system and an architecture based on fog computing and wireless
visual sensor networks is adopted. Furthermore, we propose an
approach for selecting the camera nodes that will participate in
the tracking process and we simulated three different use cases
to test the effectiveness of our system in terms of target
detection. The simulation results show that our system is a
promising solution for smart city surveillance applications.
Index Terms—Fog computing, smart city, target detection,
target tracking, video surveillance, wireless visual sensor
networks
I. INTRODUCTION
The rapid development of information and
communication technologies has allowed the
development of smart city applications, which can
improve the citizen daily life and makes urban planning
and city governance more efficient. According to [1], the
main goal of smart cities is to make citizens happier by
utilizing information technologies. However, security is a
prerequisite for happiness. In other words, a citizen will
not be happy if he doesn’t feel secure, especially in this
world where crime rates and terrorist attacks are
increased. Therefore, for a very long time, video
surveillance systems, such as Closed-circuit television
Manuscript received September 20, 2020; revised April 13, 2021.
Corresponding author email: afaf.mosaif@edu.uca.ac.ma.
doi: 10.12720/jcm.16.5.175-184
(CCTV) systems, are being used to monitor arias of
interest in the city and to deliver the collected data to a
central facility, where it is visualized and analyzed by a
human operator. However, the attention of most
individuals decreases below acceptable levels after only
20 min [2]. As a result, the live person watching the
monitor cannot detect and keep tracking all information,
especially those that are occurring at the same time in
different areas.
Therefore, Wireless Visual Sensor Networks (WVSNs)
are used in our proposed surveillance system in order to
benefit from its advantages such as: - WVSNs consist of
a large number of tiny visual sensor nodes called camera
nodes, which integrate an image sensor, an embedded
processor, and a wireless transceiver[3], - The camera
nodes can be deployed more easily without the need of
new cables installation [4], - The camera nodes can
process the collected data locally (on-board) which
reduces the total amount of data communicated through
the network and it can provide different levels of
intelligence depending on the used processing algorithms
[3], - Based on exchanged information, the camera nodes
can collaborate and reason autonomously [3] and send
just the useful information to the Base Station (BS) for
further analysis.
However, WVSNs are constrained in processing,
memory, and energy resources. Therefore, they are still
facing several issues, such as finding efficient
collaborative image processing, finding efficient coding
techniques and finding how to reliably send the relevant
visual data from the camera nodes to the aggregation
nodes or the BS in an energy-efficient way [5].
Furthermore, differing from scalar WSN nodes, the
camera nodes in WVSN are characterized by a limited
directional view, called Field-of-View (FoV), defined by
the camera direction, its angle of view, its depth of view
and its location, which affects the sensing coverage of the
network. For that reason, coverage and connectivity have
to be considered in this type of network.
To ensure public safety in the city, the camera nodes
generate every day a massive amount of data that must be
analyzed in real-time, especially in emergency situations,
which request quick response and low latency. Therefore,
Journal of Communications Vol. 16, No. 5, May 2021
©2021 Journal of Communications 175
the well-known “Cloud computing” solutions have
become inefficient for analyzing and processing
streaming video in real-time due to its associated latency
challenges, the network availability and the huge volume
of data that need to be transferred to the cloud for
decision making. Thus, fog computing, an extension of
the Cloud Computing paradigm, is a promising solution
that is used in our proposed surveillance system as it
brings the benefits of cloud computing to the edge of the
network [6]. As a result, information fusion, video
analytics algorithms and quick decisions can now be
located and performed on fog nodes close to the cameras,
allowing for faster security decisions and more reliable
image/video data transmissions. Furthermore, by using
fog computing we can benefit from the service of the fog
storage [7] and other characteristics summarized in Fig. 1
[8].
As video surveillance requires highly-reliable
connectivity and bandwidth, 5G technology can be
adopted in our proposed system as a technology of
wireless communication because it enables much lower
latency in service delivery and higher signal capacity and
speed communication compared to existing cellular
systems [9].
In summary, the main contribution of the present work
is to present a new wireless system for video surveillance
in smart cities, in which connected transportation
equipped by wireless camera nodes are used as mobile
nodes to improve the coverage of the system and an
architecture based on fog computing is used to have real-
time video surveillance, instant decision making, rapid
reaction to emergency situations and benefit from its
storage services. In addition, an approach for selecting
the camera nodes that will participate in the tracking
process is presented and 3 different use cases are
simulated to test the effectiveness of the system in terms
of target detection.
The remainder of this paper is organized as follows.
Section II presents the related works, followed by Section
III that describes in detail our proposed system, its
camera nodes operations, and the method used to select
the camera nodes that will participate in the tracking
process. Section IV presents three use cases with the
details and results of the simulation experiments. Finally,
conclusions and future works are presented in Section V.
Fig. 1. Characteristics of fog computing
II. RELATED WORKS
WVSN is used in many researches to ensure safety in
smart cities. Authors in [10] proposed a wireless system
solution that uses existing public bus transit systems in
future smart cities to collect video data from the cameras
and physically deliver it for aggregation with a maximum
median delay of 5 minutes, to be uploaded then to the
data center. In this work, the future smart city is supposed
to be equipped with video surveillance high definition
cameras in every intersection of the city to cover 360◦ of
vision in every street corner. In [4], Peixoto and Costa
presented a relevance-based algorithm to position
multiple mobile sinks in WVSN deployed on vehicles
along roads and streets of a smart city in order to collect
data from scalar sensors and cameras. The proposed
algorithm detects forbidden and disconnected zones and
dynamically computes the optimal positions of the sinks
in order to position them in permitted areas closer to
source nodes with higher sensing relevance. These
presented approaches use also vehicles as mobile nodes
of the network but as a mobile sink, not as node source
like in our proposed approach. Generally, the idea behind
using a mobile sink in WVSN is to reduce the energy
consumption of the network. In our case, the energy
preservation is not necessarily an issue as the mobile
nodes are the cameras installed and powered by vehicles
and the static nodes are powered by solar panels. Besides,
mobile sinks can be used more in delay-tolerant
applications, or our proposed system is designed for real-
time surveillance.
In order to monitor an area of interest, WVSN nodes
are deployed in a random or in a planned way to cover
this area. As explained in the introduction, the camera
nodes in WVSN are characterized by a limited directional
view (FoV) that affects the sensing coverage of the
network. Therefore, various coverage optimization
methods are proposed and presented in the literature to
maximize the network coverage while minimizing the
coverage holes and the covered overlapping areas.
In the deployment step, strategies based on Voronoi-
diagram or based on an algorithm such as Virtual Force
Algorithm are proposed in the literature to enhance the
network coverage (e.g. in [11] and [12] respectively).
Whereas, the motility (rotating the node or the lens of the
node) and/or the mobility of nodes are used in other
works such as [13] to optimize sensing coverage of
random deployed wireless multimedia sensor networks.
In our case, these methods can be applied in the static
part of our proposed system to maximize the coverage,
while the mobile part will ensure the coverage of the
public transportation interior and some uncovered areas
in the network such as the places where the mobile nodes
are stationed or rural zones in mobile nodes roads.
Since high coverage results in high tracking accuracy,
the new areas covered in our proposed system will
enhance the tracking of mobile targets in the network.
Target tracking is an important application in WVSN,
which involves the detection and localization of targets
that move in the monitored area. In the literature, the
prediction is used to activate or deactivate camera sensors
during the tracking process to preserve the energy of the
Journal of Communications Vol. 16, No. 5, May 2021
©2021 Journal of Communications 176
network such as [14]–[16]. In our case, the prediction is
used to switch sensors from mode to mode to facilitate
the tracking process and have an efficient collaboration
between the network nodes. It is very important to note
that mobile nodes in our approach don’t follow the target
like in [17], it just collaborate with the static nodes in
order to give more information about the targets and to
track it when it is detected in uncovered zones. Thus, the
system can obtain the position of the target and track it in
a map and have slices of video describing its activity.
To deal with the latency challenges caused by the
fusion and the processing of the large amount of video
data generated by the city's camera nodes in real-time,
Fog Computing is proposed as a solution in many
researches. In [6], the authors proposed an urban
speeding traffic monitoring system using the fog
computing paradigm. In which, a drone is used to monitor
the vehicles on the roads and sends the raw stream back
to the ground control station where it is displayed on a
screen. Instead of forwarding the whole video frame, the
sub-area including the suspicious vehicle is identified by
the human operator and extracted from the original frame
and sent then to the Fog Computing node where a
dynamic, real-time tracking algorithm is executed.
Finally, the tracking result is sent back to the end-users.
Authors in [18], presented the hardware options and key
software components of their novel proposed architecture
for a smart surveillance system, which is based on edge
and fog computing concepts. The Edge computing is
realized by a camera embedded system while for the fog
computing a Video Content Management Software was
developed for processing and logging of
multimedia/heterogeneous content.
In our proposed system, an architecture based on fog
computing concept is used to have on-site and instant
decision making to manage emergency situations. Fog
Computing nodes in our architecture not only provides
the computing resource but also the storage space to store
the mobile nodes collected data.
III. OUR PROPOSED SYSTEM
A. System Description
We propose in this paper a new system that aims to
have real-time video surveillance in smart cities by using
wireless camera nodes connected to fog computing nodes
and the Cloud. As shown in Fig. 2, the system is
composed of hybrid nodes: static and mobile ones.
The static camera nodes are organized in static cluster-
based architecture, where the network is divided into
smaller groups called Clusters. In each Cluster, there are
Cluster Members (camera nodes) and Cluster Heads (fog
nodes).
Fig. 2. The proposed system
The Cluster Members (CMs) detect important
objects/events and send the data to their Cluster Heads for
further processing (e.g. recognition, tracking, information
fusion, etc.). The Cluster Heads (CHs) can exchange
information with other neighboring CHs and they serve
as relays for transmitting the data to the BS (Cloud).
The mobile camera nodes of our surveillance system
are presented by the wireless camera nodes installed in
public or private transportation like buses, taxis, etc., and
are connected to the system network. Therefore, these
camera nodes present the mobile part of the system and
each one presents, temporary, a mobile CM of a CH in
the static part of the system. As shown in Fig. 3, the
vehicles can be equipped with one or more wireless
camera sensors that record from the vehicles’ exterior
and/or interior. For example, Fig. 3 (a) shows a car
Journal of Communications Vol. 16, No. 5, May 2021
©2021 Journal of Communications 177
equipped with a node that has 2 camera sensors, one of
which records in the front and the other in the rear.
The use of mobile camera nodes in our proposed video
surveillance system enhances the coverage by monitoring
non-covered areas. Therefore, isolated regions are related
and the connectivity constraint of the network is relaxed.
Also, it enhances the target-tracking application by giving
a close-up view of the target in order to have more
information about it.
Fig. 3. Types of vehicles with attached camera nodes
To sum-up, the camera nodes of our system are
composed of Static Cluster Members (SCMs) and Mobile
Cluster Members (MCMs). The SCMs are composed of
the static WVSN nodes. The MCMs are composed of
mobile WVSN nodes classified into 3 types are: - Mobile
nodes with a known trajectory, such as public buses. -
Mobile nodes with random mobility, such as cars. -
Mobile nodes with a long time of stops, such as food
trucks.
B. System Architecture
Fig. 4 presents our proposed system architecture,
which consists of three units: a data collection unit
composed of static and mobile cluster members, a fog
computing unit composed of static Cluster Heads that
represent the fog nodes, and a remote cloud center and
data analysis unit. The network computation and storage
capabilities are distributed through the three units.
Fig. 4. Proposed system architecture
The data collection unit is responsible for image/video
data collection. The SCMs in this unit are responsible just
for lightweight video analytics (e.g. intrusion detection)
because there are always connected to the CHs where
complex video analytics can be done. Unlike the SCMs,
the MCMs can be not connected all the time to the
network (e.g. in rural zones). Therefore, they are
powerful than the SCM in terms of computation and
storage and are equipped with some complex algorithm
(e.g. object recognition) that can be used even if the
camera is out of the connected area of the city.
Instead of communicating directly to servers in the
cloud, SCMs and MCMs can communicate to an
intermediate layer represented in the present architecture
as the fog computing unit, which is responsible for on-site
data processing, information fusion, and instant decision
making.
As useful video content collected by cameras is sparse,
fog computing nodes send just relevant and filtered
frames to the cloud unit for further analysis. In addition,
these nodes are presented as the static CHs in the present
architecture and not only provide the computing resource,
but also the storage space.
Remote cloud center and data analysis unit is
responsible for long-term analysis that requires powerful
computation capability to extract valuable information
that can be used to enhance the city security and to build
historical records.
C. Assumptions
Throughout this paper, we make the following
assumptions:
Cluster Head nodes have great processing power
Each camera node in the system knows its coordinates
using Global Positioning System (GPS) or other
localization methods
Each camera node in the system can return the exact
location of every object in its field of view
Journal of Communications Vol. 16, No. 5, May 2021
©2021 Journal of Communications 178
Mobile camera nodes have more storage and
processing power than the static ones
All mobile camera nodes have the same proprieties
such as video quality, field of view, etc.
To deal with the energy problem in WVSN nodes, the
SCMs are powered by solar panels and the mobile
camera nodes are powered by batteries and
rechargeable by the vehicles
D. Camera Nodes Operations
The Static Cluster Members
As the system is for real-time video surveillance, the
SCMs cannot be in sleep mode to not lose targets and
record special events. The authors in [6], declared that the
video processing time at the Fog nodes is an issue for
real-time processing. Therefore, the output frame rate
should be equal to or higher than the input frame rate to
meet the real-time video stream processing requirement.
One solution mentioned by the authors is to decrease the
resolution of surveillance video, which reduces its data
size. However, this led to sacrifice the details in the video
stream which can be a big loss of information. To address
this problem and inspired by the work in the literature
[19], the SCMs in our proposed system will operate under
two modes: low-resolution mode and high-resolution
mode.
To save energy SCMs operate with low-resolution
mode and use lightweight algorithms to detect important
events (e.g. abnormal event, wanted person detection)
and send the data back to the CH for further analysis. The
SCMs switch high-resolution mode if they are requested
by the CH to participate in a tracking process or an event
monitoring.
The Mobile Cluster Members
Unlike the SCMs that use just lightweight algorithms
to detect important events, the MCMs are more powerful
in terms of computation and storage and are equipped
with some complex algorithms (e.g. faces and vehicles
recognition) that can be used even if the MCM is out of
the connected area of the city.
The MCMs operate under three modes: normal mode,
sleep mode, and real-time video surveillance mode.
In normal mode, the MCMs store locally its collected
data in groups (minimum of 2 groups) of video sequences
(with the same size) where each group has at least one
file. Each file contains the recorded video with some
additional information such as group id, file id, time of
record, vehicle position and speed.
As shown in Fig. 5, the groups are used in a circular
fashion when one group fills up, it will be switched to the
next group. If the archive mode is enabled and the MCM
is connected to the system network, the filled group will
be archived on the CH (Fog node) and its space will be
freed. With this option, the collected data will remain on
the fog even if the camera node is stolen or damaged.
Fig. 5. MCM storage operation
In the case where the archive mode is not enabled or if
the MCM is outside the connected area of the system, the
camera node will record in files one by one locally until it
runs out of space. Then, it will start from the first files
recorded, and delete the files from oldest to newest, one
by one.
If the MCM is outside the connected area, it will
automatically lock and save any video file that was
recorded when an important event was detected and it
will assign a priority to these locked files, in order to be
the first to be sent in case of a reconnection. In order to
win time, other mobile nodes that have as destination the
connected area can serve as a relay to transport a copy of
these locked files to the system.
Once the MCM is connected to the network, it sends a
periodic message with information about its position,
speed, etc., to inform the nearest CH that it will be
temporary its MCM. This message contains the
information about the vehicle Id, number of cameras in
the vehicle, the vehicle position (x; y) (that is expressed
as GPS coordinates), and the speed of the vehicle. It
sends also to its CH the data collected when it is out of
the connected area.
In the case where the MCM is connected to the
network and it detects an important event or a wanted
target, it notifies its associated CH by sending the video
sequences where the event is detected with additional
information (e.g. position, type of target, speed, etc.). If
the CH is occupied (all channels are used), a SCM in low-
resolution mode will serve as a relay to transmit the
MCM data that has a higher priority.
After five minutes of inactivity, the camera goes into
sleep mode and as soon as it detects a motion, it turns on
and starts recording in normal mode.
The real-time video surveillance mode is activated
only if the MCM is connected to the network and is
requested, by its associated CH, to participate in a
tracking process or an event monitoring. In this mode, the
MCM sends the collected data to the associated CH in
real-time (live streaming).
It is very important to note that the mobile nodes do
not follow the targets in the tracking process but they just
cooperate with the SCMs to have a multi-view point and
a close-up view of the target or to monitor the target in
the non-covered area. Fig. 6 summarizes the MCMs
operations.
Journal of Communications Vol. 16, No. 5, May 2021
©2021 Journal of Communications 179
Fig. 6. MCMs operations
The Cluster Heads
In our architecture, the CHs represent the fog nodes
and are responsible for on-site data processing,
information fusion, future target position prediction, and
decision making. These nodes can collaborate and share
data with its neighboring CHs for data aggregation and
target features handover. Also, they not only provide the
computing resource but also the storage space to keep on-
site relevant MCMs and SCMs videos data and send just
important filtered frames to the BS (Cloud unit) for
further analysis.
Fig. 7. SCM/MCM message information
Each SCM/MCM in the network sends their messages
to their CH. These messages contain the information
shown in Fig. 7 and described in Table I.
The CHs are connected to a Shared Database (SD) that
stores the list of the processed message ids, the event ids,
the resource states of each CH, the MCMs information,
and the searched targets information. The history of the
processed message-ids is important to not process the
same message twice, for example, a MCM out of the
connected zone can send a copy of its urgent messages
with another MCM and when it is connected to the
network it will send again these messages to its
associated CH. Therefore, the CH will check first if the
received messages are already processed or not by
searching their message_id in the SD to process just the
unprocessed ones. The event ids stored in the SD are also
important to have a history of all the camera nodes that
have participated in an event monitoring or a target
tracking and in which CH the data is stored. The MCMs
information stored in the SD gives the CHs the
information about the cameras installed in MCM theirs
type (records from the vehicles exterior or interior, faces
forward or behind the vehicle) and the searched targets
information.
TABLE I: SCM/MCM MESSAGE INFORMATION MEANINGS
Field Meaning
message_id message identifier (is unique)
message_class type of message: urgent, normal, etc.
node_type type of node : MCM or SCM
node_id camera node identifier (is unique)
message_time time when the message is generated at the
camera node
Other information this field differs from a message to another, for
example, it can contain information about the
event/target such as its speed and position, or
information about the speed and the position of
the MCM.
The CH can communicate with each other for efficient
resource allocation and load balancing. If there is an
insufficient resource available in the CH, it checks in the
SD the resource available in its neighbor CHs to reassign
some messages to them in order to serve the critical
messages within its deadline.
The CHs have also other functions such as choosing
the MCMs and the SCMs that will participate in a
tracking process or an event monitoring in real-time as
described in next subsection.
E. Our Proposed Approach for Camera Nodes
Selections
To track a detected wanted target or to monitor an
important event, the CH chooses the SCMs that will
switch to the high-resolution mode to participate in the
tracking or event monitoring process and selects the
MCMs candidates that will provide more information
about the event/target and its neighbors. However, if
multiple mobile cameras are present and at the same time
detect the same event/target as shown in Fig. 8, the CH
have to choose the best node that will be temporary a
MCM that operates in real-time mode to record in
streaming the event or to participate in the tracking
process.
Fig. 8. MCMs that detect at the same time the same event/target
Therefore, some parameters have to be considered
depending on the type of the target and the type of the
MCM and its mobility such as:
Journal of Communications Vol. 16, No. 5, May 2021
©2021 Journal of Communications 180
The angle of view and the direction, of the camera
node, that are required to know if the target/event is in
the FoV of the camera or not.
The type of the mobile node, which is necessary to
choose the best node to use depending on the type of
event (e.g. mobile nodes with a long time of stop as
food trucks will be the best choice to record a fixed
event/target).
The speed of the mobile node, which is useful for
knowing the best node allowing a long duration of
recording of the event.
The distance between the mobile node and the
target/event which is useful for choosing the best node
that gives a close-up view of the event.
If the target is mobile, other parameters must be
considered such as:
The next position of the target/event, which is
required to know the nodes that can participate in the
tracking process.
The speed of the target/event, which is needed to be
compared to the speed of vehicles to choose the best
one that well gives a long recording time of the event.
Type of the target/event (vehicle or pedestrian), which
can give information about the next position of the
target and it can be useful in the prediction step.
In this paper, we assume that all MCMs have the same
proprieties. Otherwise, more other parameters such as
video quality, processing power, data rate, network signal
strength, etc. need to be considered.
MCM Nodes Selection Procedure
Fig. 9. MCMs selection zones
Once the event/target is detected for the first time by a
SCM or MCM, the CH receives the information about the
type of the event/target, its speed, and its position. Then,
the CH attributes an identifier (target_id) to this
event/target, and selects a list of camera nodes that can
have a close-up view of the target/event at an instant t0
(zone1 in Fig. 9) and the instant t+t0 (zone2 in Fig. 9).
These candidates can also give more information about
the situation of the neighbor area of the event. In Fig. 9,
r1 and r2 can be adjusted basing on the position, the type
of the event/target and its mobility.
After the selection of the candidates, the CH uses our
proposed MCM election algorithm to choose from zone1
a candidate as a source node to send streams about the
event in real-time, this node is called the Elected MCM
(E-MCM).
In order to reduce system upgrades, calculation and
communication costs, the E-MCM re-election procedure
is non-periodic and is only invoked when the E-MCM no
longer meets the eligibility criteria requested by the CH
such as:
Interruption of the connection
The event is no more in the FoV of the E-MCM
An obstacle (object/structure) is blocking the E-MCM
view
E-MCM failure due to hardware failures, physical
disasters…
E-MCM camera has bad viewing conditions such as
illumination (external conditions)
At the end of the tracking or the event monitoring
process, the CH archives the list of candidates in which
the target was detected, to use it in the future to have
more information and a different view of the event.
MCM Election Algorithm
The CH choose the E-MCM based on the following
steps:
Step 1: Each candidate k in zone1 that has the event in
its FoV calculates locally (for the first time) its selection
index at an instant t using the following equations:
where : is the Euclidean distance between
the node k and the target at the instant t
: is the speed of the mobile node k at the
instant t is the speed of the target at the instant t
: takes a value 0 or 1 depends to the type of the
node k and the type of the target at the instant t, as
shown in the Table II.
TABLE II: VTYPE VALUES
Target type
Fix (stationary) mobile
Mobile node type
Fix (stationary) 0 1
mobile 1 0
Step 2: Each candidate k sends then its selection index
to the CH in the form of a message shown in Fig. 10. The
meaning of the message fields is described in Table III.
Journal of Communications Vol. 16, No. 5, May 2021
©2021 Journal of Communications 181
Fig. 10. a MCM candidate message information
TABLE III: A MCM CANDIDATE MESSAGE INFORMATION MEANINGS
Fields
Meanings
sensor_id
camera sensor identifier (is unique)
cluster_id
the Cluster Head Id
target_id
Target/event identifier (is unique)
Vspeed
vehicle speed
Vposition
and position (x; y) (that is expressed as GPS
coordinates) information
Tspeed
target speed
Tposition
target position
Ttype
target type (abnormal event or target etc)
the vehicle selection index
Step 3: The CH sort the candidates according to their
index of selection in ascending order, and the first
element in the list is chosen E-MCM to send streams
about the event in real-time.
The E-MCM re-election procedure is non-periodic.
Therefore, the candidates recalculate their index of
selection and send the message described in Fig. 10 only
when it is requested by the CH.
Algorithms 1 describes the steps running at the CH in
order to choose the E-MCM.
Algorithm 1: Algorithm running at the CH
1: Receive the Event/target location and its speed.
2: Select the MCM that has and add them
to a list of candidates.
3: Receive the calculated index of selection of each candidate k
4: Sort the candidates according to their index of selection in
ascending order
5: Choose the first candidate to be the E-MCM that will provide a
live streaming
6: Verify, in real-time, if E-MCM meets the eligibility criteria
requested by the CH
7: if E-MCM no longer meets the eligible criteria requested by the
CH:
Repeat the steps 2,3,4,5,6
IV. PERFORMANCE EVALUATION
In this section, 3 scenarios are simulated which are:
wanted person detection, wanted vehicle detection, and
abnormal event monitoring. The general simulations
parameters are summarized in Table IV. A section of
Marrakech is imported from OpenStreetMap[20] then is
used by SUMO[21] to generate vehicles paths and
movement. The vehicles movement generated by SUMO
is then imported into ns-3 simulator, using our visual
node module presented in [22], to simulate the 3
scenarios in a wireless visual sensor network.
Fig. 11 presents a screenshot of the simulation with the
target position in the 3 scenarios at an instant t different
to 0.
Fig. 11. Simulation screenshot
TABLE IV. S
Simulation time [s] 100
Road network Section of Marrakech
Number of static sensors (SCM) 49
Number of mobile sensors (MCM) 12
Angle of view of vi (2 α) [rad] π/3
Depth of view (RV ) of the sensors [m] 20
Distance between the SCM [m] 30
Number of targets 1
Radius of zone1 (r1) [m] 50
A. Use Case 1: Wanted Person Detection
Tracking a wanted person is a difficult task, which
needs the collaboration of all camera nodes to detect the
target and to track its movement positions. The proposed
system in this article permits communication between
camera nodes attached in public and private vehicles and
static surveillance cameras in the streets. Therefore,
tracking a wanted person in this system will be faster than
the traditional systems (that contains just SCM) and more
areas will be covered such as the interior of public
transportations.
In this first use case, the search for a wanted person
moves randomly in the zone of interest is simulated. The
target speed corresponds to a pedestrian (1,38m/s) and the
mobility model used is the Random Walk 2d.
The results of the simulation show that the target was
moving in an area not covered by static cameras.
However, it was detected 1 time after 10 seconds from
the beginning of the simulation by MCM which is a
camera node installed in a vehicle parked in this area.
Journal of Communications Vol. 16, No. 5, May 2021
©2021 Journal of Communications 182
IMULATION PARAMETERS
Therefore, compared to the traditional systems, there
are more chances to detect the target with our proposed
surveillance system.
B. Use Case 2: Wanted Vehicle Detection
In this second use case, the search for a wanted vehicle
is simulated. The movement of this target is generated
also by SUMO.
The results of the simulation show that the target was
moving first in an area which is not covered by SCM,
however, it was detected by 1 MCM. Then at t=23s the
target entered the covered area and was detected by
another MCM and 8 SCM. As a summary, the target is
detected by 10 nodes where 2 are MCM and 8 are SCM.
Therefore, our proposed system can be useful too in
the case of vehicle detection and tracking, as it provides a
close-up view of the target, which gives more information
about vehicle driver, vehicle direction, etc. Also, there are
more chances to detect the wanted vehicles even in non-
covered and in rural zones (by the MCM).
C. Use Case 3: Abnormal Event Monitoring (accident...)
In this third use case, a fixed event monitoring is
simulated. The results of the simulation show that the
event is detected for the first time by a SCM, which sends
its position to the network. After 19s, the event is
detected by 1 MCM (node 1). At t=27s it is detected then
by another MCM (node 11) and at t=33s the event is
detected by 2 MCM (node 1 and node15). Finally, at
t=62s it is detected by 1 new MCM (node 13). To sum up,
the event was detected by 5 nodes, 4 MCM and 1 SCM
which monitor the fix event during all the simulation time.
The MCMs in our proposed system will give more
details and a close up view of the monitored event. In
addition, the MCM candidates will give more information
about the situation of the neighbor area of the event,
which will be helpful in decision making such as which
road must be closed, etc.
V. CONCLUSIONS
In this paper, a new system of video surveillance in the
smart cities was presented. This system aims to have real-
time video surveillance, instant decision making, rapid
reaction to emergency situations and cover more zones in
the city and its suburban areas. Therefore, transportations
equipped by camera nodes are used as the mobile part of
the system and an architecture based on fog computing
and wireless visual sensor networks is adopted.
In addition, an approach for selecting the camera nodes
that will participate in the tracking process was proposed
in the current paper, and 3 different use cases are
simulated, which are: wanted person detection, wanted
vehicle detection, and abnormal event monitoring. The
results of the simulations show that there is more chance
to detect the target with our proposed surveillance system
compared to the traditional one.
The proposed system will help law enforcement to get
information quickly, deal with some emergency situations
and find missing people or wanted targets in a more
efficient way.
For future work, we will propose a load balancing
algorithm at the fog nodes to process the received
messages according to their levels of importance and
delay tolerance.
CONFLICT OF INTEREST
The authors declare no conflict of interest.
AUTHOR CONTRIBUTIONS
Mosaif conducted the research and wrote the paper
under the guidance and supervision of the Prof. Rakrak.
All authors had approved the final version.
REFERENCES
[1] C. T. Yin, Z. Xiong, H. Chen, J. Y. Wang, D. Cooper, and
B. David, “A literature survey on smart cities,” Sci. China
Inf. Sci., vol. 58, no. 10, pp. 1–18, 2015.
[2] M. W. Green, “The appropriate and effective use of
security technologies in U.S. schools: A guide for schools
and law enforcement agencies,” 1999.
[3] S. Soro and W. Heinzelman, “A survey of visual sensor
networks,” Adv. Multimed., vol. 2009, p. 21, 2009.
[4] J. P. J. Peixoto and D. G. Costa, “Wireless visual sensor
networks for smart city applications: A relevance-based
approach for multiple sinks mobility,” Futur. Gener.
Comput. Syst., vol. 76, pp. 51–62, 2017.
[5] M. M. Y. Charfi, B. Canada, and N. Wakamiya,
“Challenging issues in visual sensor networks,” IEEE
Wirel. Commun., vol. 16, no. 2, 2009.
[6] N. Chen, Y. Chen, Y. You, H. Ling, P. Liang, and R.
Zimmermann, “Dynamic urban surveillance video stream
processing using fog computing,” in Proc. IEEE 2nd Int.
Conf. Multimed. Big Data, BigMM 2016, 2016, pp. 105–
112.
[7] J. Kharel, H. T. Reda, and S. Y. Shin, “An architecture for
smart health monitoring system based on fog computing,”
J. Commun., vol. 12, no. 4, pp. 228–233, 2017.
[8] H. Atlam, R. Walters, and G. Wills, “Fog computing and
the internet of things: A review,” Big Data Cogn. Comput.,
vol. 2, no. 2, p. 10, Apr. 2018.
[9] R. Mahmud, R. Kotagiri, and R. Buyya, “Fog computing:
A taxonomy, survey and future directions,” Internet of
Everything, pp. 103–130, 2018.
[10] W. Zhou, D. Saha, and S. Rangarajan, “A system
architecture to aggregate video surveillance data in smart
cities,” in Proc. IEEE Glob. Commun. Conf. GLOBECOM
2015, 2015.
[11] T. W. Sung and C. S. Yang, “A voronoi-based sensor
handover protocol for target tracking in distributed visual
sensor networks,” Int. J. Distrib. Sens. Networks, vol.
2014, no. 1, 2014.
[12] I. Boulanouar, S. Lohier, A. Rachedi, and G. Roussel, “A
collaborative tracking algorithm for communicating target
in wireless multimedia sensor networks,” in Proc. 7th
Journal of Communications Vol. 16, No. 5, May 2021
©2021 Journal of Communications
183
IFIP Wireless and Mobile Networking Conference
(WMNC), 2014, pp. 1–7.
[13] J. Pournazari, M. Alaei, and F. Yazdanpanah, “An energy
efficient autonomous method for coverage optimization in
wireless multimedia sensor networks,” Wirel. Pers.
Commun., vol. 99, no. 2, pp. 717–736, 2018.
[14] M. Sabokrou, “Mobile target tracking in non-overlapping
wireless visual sensor networks using neural networks,” in
Proc. 3rd International Conference on Computer and
Knowledge Engineering (ICCKE 2013), 2013, pp. 1–6.
[15] J. Amudha and P. Arpita, “Multi-Camera activation
scheme for target tracking with dynamic active camera
group and virtual grid-based target recovery,” Procedia
Comput. Sci., vol. 58, pp. 241–248, 2015.
[16] I. Boulanouar, S. Lohier, A. Rachedi, and G. Roussel,
“PTA : A predictive tracking algorithm in wireless
multimedia sensor networks,” in Proc. Global
Information Infrastructure Symposium-GIIS 2013, 2013,
pp. 1–6.
[17] T. Wang, et al., “Following targets for mobile tracking in
wireless sensor networks,” ACM Trans. Sens. Networks,
vol. 12, no. 4, pp. 1–24, Sep. 2016.
[18] G. Kioumourtzis, M. Skitsas, N. Zotos, and A. Sideris,
“Wide area video surveillane based on edge and fog
computing concept,” in Proc. 8th Int. Conf. Information,
Intell. Syst. Appl. IISA 2017, vol. 2018–Janua, pp. 1–6,
2018.
[19] O. Alaoui-Fdili, P. Corlay, Y. Fakhri, F. X. Coudoux, and
D. Aboutajdine, “A Cross-layer approach for video
delivery over wireless video sensor networks,” in TarXiv
preprint arXiv:1501.07362, 2015.
[20] J. Vargas, S. Srivastava, D. Tuia, and A. Falcao,
“OpenStreetMap: Challenges and opportunities in
machine learning and remote sensing,” IEEE Geosci.
Remote Sens. Mag., pp. 1–13, Jul. 2020.
[21] M. Behrisch, L. Bieker, J. Erdmann, and D. Krajzewicz,
“SUMO - Simulation of urban mobility - an overview,” in
Proc. 3rd Int. Conf. Adv. Syst. Simul., no. c, pp. 63–68,
2011.
[22] A. Mosaif and S. Rakrak, “Visual node module: An open-
source extension to the ns-3 network simulator,” Int. J.
Commun. Syst., vol. 33, no. 12, pp. 1–18, 2020.
Copyright © 2021 by the authors. This is an open access article
distributed under the Creative Commons Attribution License
(CC BY-NC-ND 4.0), which permits use, distribution and
reproduction in any medium, provided that the article is
properly cited, the use is non-commercial and no modifications
or adaptations are made.
Mosaif Afaf is an engineer in Computer
science, Network and Information
Systems and Ph.D. student in the
computer science department at
Computer and Systems Engineering
Laboratory (L2IS), Faculty of Science
and Techniques, Cadi Ayyad University,
Marrakech, Morocco. Her main research
interests are related to Wireless Visual Sensor Networks and the
Internet of Things.
Rakrak Said is a full professor at the
Faculty of Sciences and Techniques,
Cadi Ayyad University, Marrakech,
Morocco. He is a researcher in the
computer science department and head
of the Computer and Systems
Engineering Laboratory (L2IS). His
main research interests are related to the
Internet of Things, wireless sensor networks and cloud
computing.
Journal of Communications Vol. 16, No. 5, May 2021
©2021 Journal of Communications
184