SPECIAL SECTION ON EMERGING CLOUD-BASED WIRELESS COMMUNICATIONS AND NETWORKS

Received July 24, 2015, accepted August 16, 2015, date of publication September 9, 2015, date of current version September 23, 2015.
Digital Object Identifier 10.1109/ACCESS.2015.2477299

A Cloud-Based Smart-Parking System Based

on Internet-of-Things Technologies
THANH NAM PHAM1 , MING-FONG TSAI1 , DUC BINH NGUYEN1 ,
CHYI-REN DOW1 , AND DER-JIUNN DENG2
1 Department of Information Engineering and Computer Science, Feng Chia University, Taichung 407, Taiwan
2 Department of Computer Science and Information Engineering, National Changhua University of Education, Changhua 500, Taiwan
Corresponding author: D.-J. Deng ([email protected])

ABSTRACT This paper introduces a novel algorithm that increases the efficiency of the current cloud-based
smart-parking system and develops a network architecture based on the Internet-of-Things technology. This
paper proposed a system that helps users automatically find a free parking space at the least cost based on
new performance metrics to calculate the user parking cost by considering the distance and the total number
of free places in each car park. This cost will be used to offer a solution of finding an available parking space
upon a request by the user and a solution of suggesting a new car park if the current car park is full. The
simulation results show that the algorithm helps improve the probability of successful parking and minimizes
the user waiting time. We also successfully implemented the proposed system in the real world.

INDEX TERMS Smart-parking system, performance metrics.

I. INTRODUCTION To resolve the aforementioned problems and take

In the development of traffic management systems, an intel- advantage of the significant development in technology, the
ligent parking system was created to reduce the cost of Internet-of-Things technology (IoT) has created a revolution
hiring people and for optimal use of resources for car-park in many fields in life as well as in smart-parking system (SPS)
owners. Currently, the common method of finding a parking technology [20]. The present study proposes and develops
space is manual where the driver usually finds a space in an effective cloud-based SPS solution based on the Internet
the street through luck and experience. This process takes of Things. Our system constructs each car park as an
time and effort and may lead to the worst case of failing IoT network, and the data that include the vehicle GPS loca-
to find any parking space if the driver is driving in a tion, distance between car parking areas and number of free
city with high vehicle density. The alternative is to find a slots in car park areas will be transferred to the data center.
predefined car park with high capacity. However, this is The data center serves as a cloud server to calculate the costs
not an optimal solution because the car park could usually of a parking request, and these costs are frequently updated
be far away from the user destination. In recent years, and are accessible any time by the vehicles in the network.
research has used vehicle-to-vehicle [21] and vehicle-to- The SPS is based on several innovative technologies and can
infrastructure [23] interaction with the support of various automatically monitor and manage car parks. Furthermore,
wireless network technologies such as radio frequency iden- in the proposed system, each car park can function indepen-
tification (RFID), Zigbee, wireless mess network [22], and dently as a traditional car park. This research also implements
the Internet. This study aimed to provide information about a system prototype with wireless access in an open-source
nearby parking spaces for the driver and to make a reservation physical computing platform based on Arduino with RFID
minutes earlier using supported devices such as smartphones technology using a smartphone that provides the communi-
or tablet PCs. Furthermore, the services use the ID of each cation and user interface for both the control system and the
vehicle in booking a parking space. However, the current vehicles to verify the feasibility of the proposed system.
intelligent parking system does not provide an overall optimal
solution in finding an available parking space, does not solve A. RELATED WORKS
the problem of load balancing, does not provide economic In some studies [1]–[3], the authors proposed a new algorithm
benefit, and does not plan for vehicle-refusal service. for treatment planning in real-time parking. First, they used an

2169-3536
2015 IEEE. Translations and content mining are permitted for academic research only.
VOLUME 3, 2015 Personal use is also permitted, but republication/redistribution requires IEEE permission. 1581
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 T. N. Pham et al.: Cloud-Based SPS Based on IoT Technologies

algorithm to schedule the online problem of a parking system and implemented an SPS [7] to solve the parking problem.
into an offline problem. Second, they set up a mathematical A part of this system is implemented in the Zigbee network
model describing the offline problem as a linear problem. which sends urgent information to a PC through a coordinator
Third, they designed an algorithm to solve this linear and then updates the database. The application layer can
problem. Finally, they evaluated the proposed algorithm using quickly pass the parking information over the Internet, and
experimental simulations of the system. The experimental use the advantages of a web service to gather all the scattered
results indicated timely and efficient performance. However, parking information for the convenience of those who want to
these papers do not mention the resource reservation mecha- find a parking space. This paper simply reports the design and
nism (all parking requirements are derived immediately and implementation of an SPS and does not evaluate the system
are placed in the queue), the mechanism for assessing the performance.
resources system, the mechanism to guide vehicles to the Bonde et al. [24] aimed to automate the car and the car
parking space, the mechanism for handling situations when parking. The paper discusses a project which presents a
the request for service is denied and do not calculate the miniature model of an automated car parking system that can
average waiting time and average total time that each vehicle regulate and manage the number of cars that can be parked
spends on the system. in a given area at any given time based on the availability
In another study [4], the authors propose an SPS based on of parking spaces. The automated parking method allows
the integration of UHF frequency, RFID and IEEE 802.15.4 the parking and exiting of cars using sensing devices. Entry
Wireless Sensor Network technologies. This system can to or exit from the car park is commanded by an Android
collect information about the state of occupancy of the car based application. The difference between the Bonde system
parks, and can direct drivers to the nearest vacant parking and the other existing systems is that the authors were
spot by using a software application. However, in this work, aiming to make the system as little human dependent as
the authors have no mathematical equations for the system possible by automating the cars as well as the entire car
architecture and do not create a large-scale parking system. park; on the other hand, most existing systems require
The results of this paper only implement the proposed archi- human intervention (the car owner or other) to park the car.
tecture; they do not mention the performance of the parking Lambrinos and Dosis [19] described a new SPS architecture
system. Hsu et al. [5] proposed an innovative system includ- based on the Internet of Things technology. The architec-
ing the parking guidance service. A parking space can be ture of this system consists of a Zigbee Wireless Sensor
reserved by a smartphone via Internet access. Upon entering Network (WSN), an IoT middleware layer and a front-end
the car park, the reserved parking space will be displayed on layer as the final user interface that provides data reporting to
a small map using wireless transmission for vehicles under the user. However, there are disadvantage as it does not use a
the dedicated short-range communication protocol DSRC. suitable application protocol for the transfer of data from the
An inertial navigation system (INS) is implemented to guide WSN to the server, such as the constrained application pro-
the vehicle to the reserved space. The system will periodically tocol (CoAP), there is no mathematical model for the system
update the status of the parking space in real time to help operations, and there is no system performance evaluation.
ensure system accuracy. System performance is measured
through the accuracy of the inertial navigation systems run B. CONTRIBUTIONS
in an indoor environment, and the system implementation is With the aim of overcoming the disadvantages of the systems
evaluated by considering the accuracy of the GPS. In this mentioned above and inspired by [1]–[4] and some relevant
paper, the authors have not evaluated the performance of the works [17], [18], we introduce new SPS architecture based on
parking services, they do not provide any mathematical model IoT and build a mathematical model of the system operation.
of the system, and do not consider the waiting time of each First, our algorithm adopts a mechanism to search car parks at
vehicle for service. the least cost. Second, we adopt a mechanism for forwarding
Other researchers have designed architecture for parking the vehicles to another car park if the current car park is
management in smart cities [6]. They proposed intelligent full. We propose a network of car parks such that each park
parking assistant (IPA) architecture aimed at overcoming is a node in a network. Each node obtains the information
current public parking management solutions. This architec- from the neighboring node, thus ensuring smooth movement
ture provides drivers with information about on-street parking of vehicles at low cost and increasing the probability of
stall availability and allow drivers to reserve the most conve- finding a free parking space. Our system achieves better
nient parking stall at their destination before their departure. performance compared with other parking systems. We eval-
They use RFID technology in this system. When a car parks uated the performance of our system through simulation and
or leaves the IPA parking spot, the RFID reader and the implementation. The results of the simulation are close to our
magnetic loop detect the action and send this information to mathematical models and achieve better performance than the
the unit controller to update the information on the car park other systems. The proposed system reduces the number of
status. This study uses only some simple mathematical equa- vehicles failing to find a parking space and minimizes the
tions for the system architecture and does not create a large- costs of moving to the car park. The cost defined here is the
scale parking system. In other works, authors have designed time that the user must wait for the service, thus helping users

1582 VOLUME 3, 2015

T. N. Pham et al.: Cloud-Based SPS Based on IoT Technologies

save time and money and reducing environmental pollution. at each car park. The system allows a driver to search
We have also successfully implemented our system in a uni- and find information on parking spaces from each car
versity parking system. park without the need to directly access the local server
node by directly accessing the cloud-based server.
C. ORGANIZATION
This paper is organized as follows: Section II describes
the proposed architecture including the system and network
architectures. Section III discusses the algorithms and the
mathematical models of the system. Section IV presents the
simulation. Section V is the implementation of the system.
Section VI is our conclusion and suggestions for future work.

II. PROPOSED ARCHITECTURE

A. SYSTEM OVERVIEW FIGURE 2. Local unit.
The system is derived from the idea of IoT [13], [14]. The
system uses the WSN [15] consisting of RFID technology • Local Unit: This unit is located in each car park and
to monitor car parks. An RFID reader counts the percentage stores the information of each parking space, as shown
of free parking spaces in each car park. The use of RFID in Fig.2. The local unit includes the following:
facilitates implementation of a large-scale system at low cost. ◦ Control Unit: This is an Arduino module, which is
The system provides a mechanism to prevent disputes in the connected using an RFID reader. The card reader
car park and helps minimize wasted time in looking for a authenticates the user information and then displays
parking space. After logging into the system, the user can this information on the screen. If the information
choose a suitable parking space. Information on the selected of the RFID tag or card is correct, the Arduino
parking location will be confirmed to the user via notification. module will control the opening of the door for the
Then, the system updates the status of the parking space to vehicle to enter. The Arduino module connects with
‘‘pending’’ during which time the system will not allow other the cloud server through an Internet connection to
users to reserve it. If after a certain period of pending time transfer data from the local car park to the cloud
the system determines that no car is parked in that space, then server database.
it changes the status to ‘‘available.’’ The system will update ◦ Screen: This displays information on the capacity
the status from the WSN node (the status of car park spaces) of the local car park, the total current percentage
when a new car joins in the system. Therefore, the status of of free spaces, the status of the RFID tag check, the
the overall parking system is always updated in real time. The user card when entering, and a mini map of the local
system will help plot the parking time for each parking space car park.
in real time and can support the business with hourly parking ◦ RFID Tag or ID Card: This is used to check
charges. and authenticate user information and calculate the
percentage of total free spaces in each car park.
• Software Client: This is an application software system.
Running on Android operating system, the users will
install it on their smartphones and use it to reserve
parking spaces. The users access the system via
3G/4G mobile connections.

C. NETWORK ARCHITECTURE
In general, we will use the term ‘‘user’’ when referring to the
driver or vehicle and the term ‘‘resources’’ when referring to
the parking spaces.

1) PARKING NETWORK
We use the car park network (CPN) architecture infrastruc-
FIGURE 1. Architecture of the proposed system. ture/backbone. The architecture is shown in Fig. 3(a), where
the dashed lines indicate wireless link and the solid lines
B. SYSTEM ARCHITECTURE indicates wired link. This type of parking network includes
Fig. 1 shows our smart IoT parking system. routers that form as the infrastructure for connected clients.
Elements in the system: The CPN infrastructure/backbone can be built to allow sensor
• Cloud-Based Server: This is a Web entity that stores networks to connect using wireless radio technologies. The
the resource information provided by local units located routers form a self-configuring and self-healing link network.

VOLUME 3, 2015 1583

 T. N. Pham et al.: Cloud-Based SPS Based on IoT Technologies

FIGURE 4. Parking network.

N1 = 100 spaces, N2 = 120 spaces, N3 = 200 spaces,

N4 = 100 spaces, N5 = 120 spaces, N6 = 120 spaces,
N7 = 100 spaces; D12 = 1.2 km, D13 = 1.6 km,
D23 = 2.0 km, D27 = 1 km, D34 = 1.5 km, D37 = 1.8 km,
D45 = 1.2 km, D56 = 0.8 km and D67 = 1.2 km. These
parameters are shown in Fig. 5 using simple neighbor tables.
In Fig. 5, we assume that the total free spaces in N1 = 20,
FIGURE 3. (a) Infrastructure/backbone of the CPN architecture.
in N2 = 60, in N3 = 60, in N4 = 70, in N5 = 60, N6 = 30
(b) CPN deployment for car parking system. and in N7 = 60. To increase the performance of finding a free
parking resource, the neighbor table in each node contains
information on the current number of free parking resources
Routers can be connected to the Internet by gateway in the neighboring nodes. Our idea is to use the number of
functionality. This approach, also referred to as infrastructure total free parking resources in each node to calculate the cost
meshing, provides the backbone for conventional clients and for choosing a car park.
enables integration of CPNs with existing WSNs through
gateway/bridge functionalities in the routers. Conventional
clients with the same radio technologies as the routers can
directly communicate with the routers.
We have assumed that each car park is a node in a CPN.
The deployment network in a real environment is shown
in Fig. 3(b) where each car park is labeled.
• P1 is car park number 1; N1 is the total parking spaces
in P1 .
• P2 is car park number 2, N2 is the total parking spaces
in P2 .
• Pn is car park number n, Nn is the total parking spaces
in Pn . FIGURE 5. Simple neighbor tables.

The total capacity of the system is N = N1 + N2 + N3 +

. . . + Nn (spaces). D is the real distance between two nodes
2) CONSTRUCTING THE NEIGHBOR TABLE OF NODES
in the network. Dij is the distance between nodes Pi and Pj .
We use a function named F(α, β) to calculate the cost
Fig. 4 shows our network.
between the nodes in the network. F(α, β) is a function that
Each node has a neighbor table to maintain information on
depends on the distance between two nodes and the number
the current status of the network and a queue with predefined
of free parking spaces in the destination node. F(α, β) is
length. The neighbor table for each node contains information
considered to be a weighted link between two nodes in the
on the neighboring nodes directly linked to it. On the other
parking network. If two nodes are not directly linked, then
hand, the queue is used to control the number of vehicles
F(α, β) = ∞. If the vehicle comes into a node and that node
forwarded to the node, which aims to prevent overloading
is full, the vehicle will be forwarded to the next node, which
in the number of vehicles beyond the capacity of the node.
is a neighbor of this node with the smallest value of F(α, β)
In our proposed system, each node will broadcast a message
in the neighbor table. We calculate the cost function F(α, β)
to its neighboring nodes after a new node joins or leaves it.
from node Pi to node Pj , i.e.,
This message includes information on its total free resources.
The neighboring node that receives this message will update dij tj
Fij = Fij (α, β) = α × +β × (1)
its neighbor tables. We have assumed that, in our network, Dup Tup
1584 VOLUME 3, 2015
T. N. Pham et al.: Cloud-Based SPS Based on IoT Technologies

where α is a coefficient that depends on the length of the path III. ALGORITHM AND MATHERMATICAL MODEL
between two nodes and β is a coefficient that depends on A. ALGORITHM
the number of free slots in the destination node. F(α, β) is We propose an algorithm to describe the operation of the
inversely proportional to the distance between two nodes and system.
directly proportional to the total free slots in the destination
node. Depending on which parameter we consider to be 1) SYSTEM OPERATIONS
the more important of the two parameters, i.e., the distance When a user wants to find a parking slot, he must login to our
or the free slots, we can adjust α and β to achieve better system. After successful login, a request message is sent to
network performance. α and β are parameters derived from search for a free parking slot. Then, the system will send back
the experiment, and their value is [0, 1]. If α = 0, we only a response message containing the information, including the
consider the number of free spaces to calculate the car park address and the directions to reach it. The choice
cost to the user. If β = 0, we only consider the dis- of the car park is based on the function F(α, β), which is
tance between two nodes to calculate the cost to the calculated based on the current location of the vehicle and the
user. location of the car park. The system will forward the vehicle
In Eq. (1), we calculate the cost function based on the to a car park with a minimum F(α, β) value if the current
distance between two nodes and the percentage of free car park is full. When the user arrives at the car park, he
parking spaces at each node. We use the upper bound must be authorized to enter. This authorization is achieved
of the distance between two nodes and the upper bound via the RFID technology or by scanning the user card. This
of the capacity for parking in each car park. In Eq. (1), mechanism is simple but economical. If the information is
dij is the distance between nodes Pi and Pj , Dup is the upper correct, the user is allowed to park. If the current car park is
bound of the distance and is a global parameter, tj is the full, the system will send a suggestion message that includes
number of spaces that are occupied at node Pj , and Tup is the information on a new car park, including the address and
upper bound of the capacity of the overall parking network new directions, with a minimum cost. The new car park
and is a global parameter. We assume a network with will be selected based on the neighbor table of the current
seven nodes as in Fig. 5 and calculate the value car park (the first node in the neighbor table), as shown
of function F with α = 0.2, β = 0.8, D = 2 km, in Fig. 7.
T = 200 spaces: F12 = 0.36; F13 = 0.72; F21 = 0.44;
F23 = 0.76; F27 = 0.34; F31 = 0.48; F32 = 0.44;
F34 = 0.27; F37 = 0.42; F43 = 0.71; F45 = 0.36;
F54 = 0.24; F56 = 0.44; F65 = 0.32; F67 = 0.36;
F72 = 0.34; F73 = 0.74; F76 = 0.48. The neighbor table
of each node with the F(α, β) function is shown in Fig. 6.
Fig. 6 shows that the new neighbor table for each node follows
Eq. (1). We will use this routing table in choosing the next
node where to forward the user when a car park is full.

FIGURE 6. Neighbor tables sorted by descending values of F(α, β). FIGURE 7. Algorithm of the system operations.

VOLUME 3, 2015 1585

 T. N. Pham et al.: Cloud-Based SPS Based on IoT Technologies

the neighboring tables of its neighboring nodes, as shown

in Fig. 9.

FIGURE 8. Vehicle processes.

Our proposed system involves two processes: reservation

and entering.
• Reservation Process: Starting from (1) to (3) shown
in Fig. 8, if the user is looking for a free parking space,
he will send a request message to the system (1), which
is done using a smartphone. When the system receives
this request, it will find car park P1 with the least cost FIGURE 9. Algorithm for updating the status of the car park.

[minimum value of F(α, β)] and forward this message to

the user. In this case, the least cost is the minimum value
of function F(α, β). The value of F(α, β) is calculated as B. MATHEMATICAL MODELS
the distance (between the vehicle and car parks) and the We build the mathematical models of our proposed system
number of free spaces in each car park. If this car park based on the results in [1] and [2]. We create a parking
has free parking slots, it will send a response message to planning strategy. We let P denote the set of all vehicles with
the user (3). The response message includes the address parking queries in the queue. We let S denote the total of all
of car park P1 and its directions. Because we use the available car parks. We let W denote the set of wij , where wij
percentage of total free spaces in suggesting a new car is the cost between vehicle pi (pi ∈ P) and car park Sj (Sj ∈ S).
park, a high probability of success exists in finding a free We can achieve W by calculating the distance from the vehi-
parking space. cle to the car park (GPS address) and the number of free
• Entering Process: Starting from (4) going to (5), if a spaces in car park Sj . We let M and N be the size of P and S,
user enters car park P1, he must be authorized using an respectively. Therefore, the size of W is M × N. By assuming
ID or an RFID card (4). If authorized, the door is opened, that vehicles are jobs and parking places are servers, Wij is
and the count will increase by one. The system will the cost for server Sj to do job Pi . We save the solution in X,
send a response message to the user to notify successful where xij ∈ X , i.e.,
parking (5). If the car park is currently full, it will send (
a response message suggesting an alternative car park, 1, if Pi will park at Sj
xij = (2)
including relevant information on new car park P2 , with 0, if Pi will not park at Sj .
the least cost.
We let C be the total cost for all vehicles in P to go to the
2) CALCULATING THE TOTAL FREE PARKING SPACES AND parking places assigned to them by the SPS, i.e.,
UPDATING THE NEIGHBOR TABLE
M X
N
In our proposed system, we use RFID technology to calculate C=
X
wij × xij . (3)
the percentage of total free parking spaces in each car park.
i=1 j=1
In each car park, an RFID reader is installed at the entrance.
We use a variable named ‘‘Count’’ to calculate the total In our study, we use F(α, β) as the cost; thus, we have a
number of vehicles in the car park. Count = Count + 1 new total cost.
when a vehicle enters, and Count = Count −1 when a vehicle M X
N
leaves. When Count = Ni , car park i is full. The process of C=
X
Fij (α, β) × xij (30 )
updating the neighbor table is described as follows: when a i=1 j=1
change in the value in the Counter occurs, which changes
the percentage of the total free parking spaces at this node, To decrease the cost to the user, we will choose the mini-
this node will send a message containing updated information mum value of F(α, β) in (30 ). We aim to make C minimum
to the cloud-based server. The cloud-based server will update on the condition that each vehicle obtains exactly one parking

1586 VOLUME 3, 2015

T. N. Pham et al.: Cloud-Based SPS Based on IoT Technologies

resource and each car park space can be assigned to only one is expressed as
vehicle, i.e.,
µ2s
Ta = . (6)
µs − µA

 PN
 xij ≤ 1


j=1 The average waiting time in the queue (expected from a
(4)
 PM long run) in the M/M/k queue is expressed as
 xij = 1.


i=1 µ2s
Ta = . (7)
N
k(µs − µA )
P
xij ≤ 1 indicates that any user in the queue may be The average waiting time of the system is
j=1
assigned at most one car park but may also fail to get an N
P
PM Tai
assignment. On the other hand, xij = 1 still guarantees Ta =
i=1
, (8)
i=1 N
that each user in the queue maintains a car park assignment.
where N is the total number of parking spaces.
In our proposed system, if a vehicle does not find a free
parking space on arrival at a full car park, forwarding it to IV. SIMULATION
a different car park will be suggested. We let h denote the A. SIMULATIONS
number of forwarded vehicles. Each car park can be assigned 1) SETUP
n
To evaluate the performance of the processes, we simulated
P
to kj vehicles (kj is the total free slot in Sj ; kj = N ), i.e.,
j=1 a network deployment, including the car park architecture
 mentioned above. We used the network simulation tool Arena
N
P x ≤ h
 to simulate this network. To simulate the mathematical and
 ij

j=1 queuing models, we randomly created vehicles to join the
(5)
M network. The arrival process of the vehicles followed the
 xij = kj .

 P
 Poisson distribution in our simulation, which was denoted
i=1
as POIS(X), where X is the inter-arrival time between suc-
The time complexity of our algorithm is O(n∗ k). We will cessive vehicles arriving at the car park. In this simulation
try to reduce the time cost for each vehicle in finding a free X = 15 and 20 min. We considered the vehicle as the job
parking resource. Our mathematical model has reduced the and the parking space as the entity doing the job. The time
total cost and reaches a better solution in distributing the users for doing the job followed an exponential distribution, which
to the overall network resources. was denoted as EXPO(Y) in this simulation, where Y is the
average service time that a vehicle stays in the parking space.
C. QUEUE MODELS We chose Y = 60 min in this case. We simulated a parking
We modeled the system into a service queue. It includes all network with five car parks as five nodes. We assumed that
users entering each car park. The entering process at each the network nodes are interconnected, as shown in Fig. 11.
node is considered to be a first-in first out (FIFO) queue and We set up each car park with a capacity for four parking
a Markov process, as shown in Fig. 10. spaces as the resources. We also created the same number of
random vehicles to arrive at each car park. In this simulation,
the number of vehicles arriving at each car park is 60, 70, 80,
90, and 100. We ran the simulations until all vehicles were
serviced.

FIGURE 10. Service queue.

The mathematical model for the entering process can be

described as follows: M/M/1/K/FIFO. The first ‘‘M’’ denotes
that the distribution of the arrival process is Markovian
(Poisson distribution), the second ‘‘M’’ stands for the service-
FIGURE 11. A five-node network.
time distribution, which is also Markovian (exponential dis-
tribution), ‘‘1’’ denotes the server, and K is the number of To compare the network performance and provide an opti-
lots. µA is the inter-arrival time between two users, and µS is mal solution for our proposed network, we set up a simu-
the service time (parking time). By assuming that µS > µA , lation based on various values of α and β. We simulated
the queue does not explode. The average waiting time in all cases of alpha and beta in the range from zero to one.
the queue (expected from a long time) in the M/M/1 queue The following are some outstanding values of α and β in the

VOLUME 3, 2015 1587

 T. N. Pham et al.: Cloud-Based SPS Based on IoT Technologies

range from zero to one. The value of alpha changed from of the algorithm that we proposed with an existing parking
α = {0, 0.2, 0.5, 0.8, 1}. The value of beta changed from system, we used two different models for evaluation. In the
β = {0, 0.2, 0.5, 0.8, 1}. We set up the distance between the first network model, as shown in Fig. 12(a), we implemented
network nodes as follows: D12 = 0.6 km, D15 = 1.2 km, the greedy method. In this method, when vehicles arrive at
D23 = 2.4 km, D24 = 2.0 km, D34 = 1.8 km, and a full car park, they will be placed in a queue and wait for
D45 = 1.6 km. We chose the value of the upper bound of the service until this car park has a free parking space. This
the distance as Dup = 2.4 km and that of the upper bound queue is the FIFO queue. The greedy method is a common
of the capacity as Tup = four spaces. All set-up parameters method that represents the traditional parking system with no
of the simulation are summarized in Table 1. planning to solve this problem. A loop is used to loop vehicles
until the node has an available parking space.
TABLE 1. Simulation parameters.

2) ARENA SIMULATION TOOL

Arena is discrete event simulation and automation software
developed by Systems Modeling and acquired by Rockwell FIGURE 12. (a) Network model with the greedy method. (b) Network
model with forwarding.
Automation in 2000 [9]–[12]. It uses the SIMAN proces-
sor and simulation language. Arena is good software that
simulates many types of real-time systems such as parking To reduce the waiting time of vehicles in the system, we use
systems. The basic building blocks of Arena models are the second network model in planning to solve this problem.
modules. Flowcharts and data objects define the process This network model is shown in Fig. 12(b). In this network
to be simulated and are chosen from panels in the project model, when a vehicle arrives at a car park that is currently
bar. Flowchart modules describe the dynamic processes in full, it will be forwarded to a different car park that has free
the model. The types of flowchart modules available are parking spaces. The forwarding is based on the algorithms
Create, Dispose, Process, Decide, Batch, Separate, Assign, that we have proposed. We simulated two network models
and Record. Other panels may contain many additional types in the Arena simulator, and we compared the average waiting
of flowchart module. Data modules define the characteristics time and average total time that a vehicle resides in each node.
of various process elements such as entities, resources, and
queues. They can also set up variables and other types of B. RESULTS AND EVALUATION
numerical values and expressions that pertain to the whole To evaluate the performance of the proposed system, we
model. determined the parameter for system performance as the cost
In particular, in the parking system simulation, Arena in terms of user time in the system. The cost to the user is the
supports many random distributions of arrival and service time that the user spends in the parking system for service.
processes, such as Poisson, Normal, Exponential, Triangular, If this cost can be minimized, we can reduce the other costs
Uniform, Beta, Gamma, Logarithmic, and Weibull distribu- such as monetary, fuel, and environmental pollution costs.
tions. It allows statistical calculations and exports the average The time in this study is the average waiting time for the
values of the parameters used to evaluate the performance, service to the user and the average total time of the user in
such as the average vehicle waiting time for parking requests the system, including the waiting, travel, and service times.
and average time the vehicle stays in the parking system. A smaller cost value leads to better system performance.
Many previous researchers have shown that the simulation Given the parameters we simulated, the parameter with the
results in Arena are close to those in actual practice. Owing smallest time cost value will be considered as the optimal
to the abovementioned advantages, we chose Arena as the solution and is used as a proposal to deploy a similar model in
simulation tool in this study. practice. Fig. 13 shows our comparison of the average waiting
time in a normal network with a loop and our proposed
3) SCENARIOS network. In the simulation, we used 50, 60, 70, 80, 90, and
As mentioned in Section IV-(1), we created a simulation net- 100 vehicles that arrived at each node. The distribution of
work consisting of five nodes. To compare the performance the inter-arrival time = POIS (15 min), which means that

1588 VOLUME 3, 2015

T. N. Pham et al.: Cloud-Based SPS Based on IoT Technologies

that of β is 0.2, our proposed network achieves the best

performance compared with the other (α, β) pairs. If the value
of α is 1 and β is 0, the average total time is approximately
equal to the average total time in a normal network, which
is the worst case. The best solution in this network with
(α = 0.2 and β = 0.8) reduces the average total time the user
stays in the system by approximately 50%. Fig. 15 shows our
simulation of 50, 60, 70, 80, 90, and 100 vehicles arriving
at each node. The distribution of the inter-arrival time =
POIS(20 min). The results show that our algorithm achieves
better performance than the system with no parking planning.
If the value of α is 0.8 and that of β is 0.2, our proposed
network realizes the best performance in the range from
60 to 70 vehicles arriving at each car park. In the range from
70 to 90 vehicles arriving at each node, the pair (α = 0 and
FIGURE 13. Average waiting time (15 min). β = 1) realizes the best performance. We can explain this
result by the fact that the number of parking spaces at each
four vehicles arrive at each car park per hour. The results network node is only four, and the arrival rate of the users
show that our algorithm achieves better performance than the decreases (20 versus 15 min); thus, the percentage of free
network model without planning. We can see that if the value parking-space factor is more meaningful than the distance.
of α is 0.8 and the value of β is 0.2, our proposed network If the value of (α = 1 and β = 0) is still the worst case, the
achieves the best performance with minimum waiting time. system experiences the longest average waiting time.
If the value of α is 1 and β is 0, our proposed network has the
longest average waiting time, which is the worst case because
we only use the distance parameter to calculate F(α, β).
If the user is only forwarded to the car park with the shortest
distance, a high probability exists that at the next car park,
the user will still not find a free parking space because the
percentage of available parking spaces is not taken into
account. The network performance in this case is not equiv-
alent to a normal network. We realize that if we use the
percentage of free spaces in each car park as a parameter for
planning with regard to forwarding the users, the waiting time
of the user for the service will be greatly reduced compared
with that in an ordinary network.

FIGURE 15. Average waiting time (20 min).

In this case, the expected inter-arrival time is 20 min, which

is longer than 15 min. Thus, Fig. 15 shows that the average
waiting time will be significantly reduced (approximately
10 times for the best case where α = 0.2 and β = 0.8)
compared with that shown in Fig. 13. The simple explanation
is that if more vehicles come into the system per hour, the
greater is the waiting time for service in terms of the total
number of parking spaces in each node, which do not change.
Fig. 16 shows the average total time of each vehicle in the nor-
mal network versus the proposed network in the case where
the distribution of the arrival process is POIS(20 min). If the
FIGURE 14. Average total time (15 min). value of α is 0.8 and that of β is 0.2, our proposed network
realizes the best performance (a minimum of the average total
Fig. 14 shows our comparison of the average total time of time). Based on the results of the simulation, we can conclude
each vehicle in a normal network model and our proposed that if we only use the distance parameter in planning for
network model. We can see that if the value of α is 0.8 and parking, the network performance will be lower than that

VOLUME 3, 2015 1589

 T. N. Pham et al.: Cloud-Based SPS Based on IoT Technologies

FIGURE 17. Implementation of the software system. (a) The login

interface; (b) the map of all distributed car parks; (c) the result of the
shortest case; (d) the result of our algorithm.
FIGURE 16. Average total time (20 min).

of the normal network. However, if we use the percentage of

total free spaces in combination with the distance parameter
in planning for parking, the network performance will signif-
icantly improve compared with that of the normal network.
In all cases, the best network performance is achieved with
the (α, β) pair of (0.2, 0.8).
V. IMPLEMENTATION
A. SOFTWARE SYSTEM
We designed a software client that runs on a smartphone
based on the Android platform [8], which was built from the
ground up to enable developers to create compelling mobile FIGURE 18. Implementation of elements in the system. (a) Implement of
applications that take full advantage of all that a handset local unit; (b) the screen display information of RFID tags; (c) the
can offer. In this phase, we use the Android SDK Tools, implementation of the cloud-based server.

Revision 24.3.4, which is a set of development tools used to

develop applications for Android platform that can be used
to write Android programs in the command prompt. The
RFID antenna, Arduino Uno R3, Arduino Ethernet Shield,
most common method is using an integrated development
Screen and Cloud-based Server system. Fig. 18(a) shows
environment. In our ideal concept, users who want to use our
the implementation of the local unit; Fig. 18(b) describes
system must be registered as a member of the system. Our
the screen display information of the RFID tags; Fig. 18(c)
cloud-based server is implemented on Apache Hadoop 2.7.1,
describes the implementation of the system cloud-based
and we use Apache HBase as our database. Apache HBase is a
server.
Hadoop database, a distributed, scalable, and large data store.
Fig. 17(a) shows the login interface of the system.
Fig. 17(b) is a map describing the distribution of all car parks VI. CONCLUSION
at a university. In this map, there are seven car parks with This study has proposed a parking system that improves
capacities of: P1 = 75 spaces, P2 = 30, P3 = 56, P4 = 60, performance by reducing the number of users that fail to
P5 = 80, P6 = 50, P7 = 85. Symbol S is the current find a parking space and minimizes the costs of moving
location of the user. Fig. 17(c) indicates the result returned to the parking space. Our proposed architecture and system
by the system when users search for a car park following the has been successfully simulated and implemented in a real
shortest distance. The result returned is P5 and the distance situation. The results show that our algorithm significantly
from the user to the car park is 42 m. Fig. 17(d) indicates the reduces the average waiting time of users for parking. Our
result returned by the system when users search for a car park results closely agree with those of our proposed mathematical
following our algorithms, here called the best case. The result models. The simulation of our system achieved the optimal
returned is car park P2 , the distance from the user to the car solution when most of the vehicles successfully found a free
park is 176 m. parking space. The average waiting time of each car park for
service becomes minimal, and the total time of each vehicle in
B. ELEMENTS each car park is reduced. In our future study, we will consider
Fig. 18 describes the implementation of the system the security aspects of our system as well as implement our
elements, including RFID tags, the RFID reader, the proposed system in large scales in the real world.

1590 VOLUME 3, 2015

T. N. Pham et al.: Cloud-Based SPS Based on IoT Technologies

REFERENCES THANH-NAM PHAM received the B.S. and

[1] Y. Geng and C. G. Cassandras, ‘‘A new ‘smart parking’ system based on M.S. degrees in electrical engineering from the
optimal resource allocation and reservations,’’ in Proc. 14th Int. IEEE Conf. Hanoi University of Science and Technology,
Intell. Transp. Syst. (ITSC), Oct. 2011, pp. 979–984. Vietnam, in 2010 and 2012, respectively. He is
[2] Y. Geng and C. G. Cassandras, ‘‘New ‘smart parking’ system based on currently pursuing the Ph.D. degree in infor-
resource allocation and reservations,’’ IEEE Trans. Intell. Transp. Syst., mation engineering with Feng Chia Univer-
vol. 14, no. 3, pp. 1129–1139, Sep. 2013. sity. His research interests include peer-to-peer
[3] X. Zhao, K. Zhao, and F. Hai, ‘‘An algorithm of parking planning networks, wireless sensor networks, and protocols
for smart parking system,’’ in Proc. 11th World Congr. Intell. Control for Internet of Things.
Autom. (WCICA), 2014, pp. 4965–4969.
[4] L. Mainetti, L. Palano, L. Patrono, M. L. Stefanizzi, and R. Vergallo,
‘‘Integration of RFID and WSN technologies in a smart parking system,’’
in Proc. 22nd Int. Conf. Softw., Telecommun. Comput. Netw. (SoftCOM), MING-FONG TSAI received the Ph.D. degree
2014, pp. 104–110.
from the Department of Electrical Engineer-
[5] C. W. Hsu, M. H. Shih, H. Y. Huang, Y. C. Shiue, and S. C. Huang,
ing, Institute of Computer and Communication
‘‘Verification of smart guiding system to search for parking space via
DSRC communication,’’ in Proc. 12th Int. Conf. ITS Telecommun. (ITST), Engineering, National Cheng Kung University,
2012, pp. 77–81. Taiwan, in 2011. He is currently an Assistant
[6] R. E. Barone, T. Giuffrè, S. M. Siniscalchi, M. A. Morgano, and Professor with the Department of Information
G. Tesoriere, ‘‘Architecture for parking management in smart cities,’’ IET Engineering and Computer Science, Feng Chia
Intell. Transp. Syst., vol. 8, no. 5, pp. 445–452, 2014. University, Taiwan. His current research interests
[7] C. Shiyao, W. Ming, L. Chen, and R. Na, ‘‘The research and imple- include error-control coding, multimedia commu-
ment of the intelligent parking reservation management system based on nications, and vehicular communications.
ZigBee technology,’’ in Proc. 6th Int. Conf. Meas. Technol. Mechatronics
Autom. (ICMTMA), 2014, pp. 741–744.
[8] D. J. Bonde, R. S. Shende, K. S. Gaikwad, A. S. Kedari, and A. U. Bhokre, DUC-BINH NGUYEN received the B.S. degree
‘‘Automated car parking system commanded by Android application,’’ in in information technology from the Thai Nguyen
Proc. Int. Conf. Comput. Commun. Inform. (ICCCI), 2014, pp. 1–4. University of Information and Communication
[9] J. E. Hammann and N. A. Markovitch, ‘‘Introduction to Arena [simulation Technology, Vietnam, in 2008, and the mas-
software],’’ in Proc. Winter Simulation Conf., 1995, pp. 519–523. ter’s degree in information technology from
[10] W. D. Kelton, R. Sadowski, and N. Zupick, Simulation With Arena, 6th ed.
the Manuel S. Enverga University Foundation,
New York, NY, USA: McGraw-Hill, 2014.
Philippines, in 2010. He is currently pursuing the
[11] T. Altiok and B. Melamed, Simulation Modeling and Analysis With
ARENA. Amsterdam, The Netherlands: Elsevier, 2007.
Ph.D. degree with the Department of Information
[12] M. D. Rossetti, Simulation Modeling With Arena. New York, NY, USA: Engineering and Computer Science, Feng Chia
Wiley, 2010. University, Taichung, Taiwan. He was a Lecturer
[13] J. Höller, V. Tsiatsis, C. Mulligan, S. Karnouskos, S. Avesand, and and the Manager of Network and Communication Department with the Thai
D. Boyle, From Machine-to-Machine to the Internet of Things: Introduc- Nguyen University of Information and Communication Technology, Thai
tion to a New Age of Intelligence. Amsterdam, The Netherlands: Elsevier, Nguyen, Vietnam. His current research interests include mobile computing,
2014. vehicle ad-hoc network, wireless ad-hoc networks, and Internet of Things.
[14] I. Wigmore, Internet of Things (IoT). Newton, MA, USA: TechTarget,
Jun. 2014.
[15] I. F. Akyildiz and I. H. Kasimoglu, ‘‘Wireless sensor and actor networks: CHYI-REN DOW was born in 1962. He received
Research challenges,’’ Ad Hoc Netw., vol. 2, no. 4, pp. 351–367, Oct. 2004. the B.S. and M.S. degrees in information
[16] K. Ashokkumar, R. B. Sam, and B. Arshadprabhu, ‘‘Cloud based intel- engineering from National Chiao Tung University,
ligent transport system,’’ in Proc. 2nd Int. Symp. Big Data Cloud Taiwan, in 1984 and 1988, respectively, and the
Comput. (ISBCC), vol. 50. 2015, pp. 58–63. M.S. and Ph.D. degrees in computer science from
[17] Z. Suryady, G. R. Sinniah, S. Haseeb, M. T. Siddique, and the University of Pittsburgh, PA, in 1992 and
M. F. M. Ezani, ‘‘Rapid development of smart parking system with
1994, respectively. He is currently a Professor with
cloud-based platforms,’’ in Proc. 5th Int. Conf. Inf. Commun. Technol.
the Department of Information Engineering and
Muslim World (ICT4M), 2014, pp. 1–6.
[18] Z. Ji, I. Ganchev, M. O’Droma, and X. Zhang, ‘‘A cloud-based intelligent Computer Science, Feng Chia University, Taiwan.
car parking services for smart cities,’’ in Proc. 31st URSI General Assembly His research interests include mobile computing,
Sci. Symp. (URSI GASS), Aug. 2014, pp. 1–4. ad-hoc wireless networks, agent techniques, fault tolerance, and learning
[19] L. Lambrinos and L. Dosis, ‘‘DisAssist: An Internet of Things and technology.
mobile communications platform for disabled parking space manage-
ment,’’ in Proc. IEEE Global Commun. Conf. (GLOBECOM), Dec. 2013,
pp. 2810–2815. DER-JIUNN DENG received the Ph.D. degree
[20] M. Du, J. Fang, and H. Cao, ‘‘A new solution for city parking guiding in electrical engineering from National Taiwan
based on Internet of Things and multi-level multi-agent,’’ in Proc. Int. Conf. University, in 2005. He joined the Department of
Electron., Commun. Control (ICECC), 2011, pp. 4093–4096. Computer Science and Information Engineering,
[21] C. Rhodes, W. Blewitt, C. Sharp, G. Ushaw, and G. Morgan, ‘‘Smart National Changhua University of Education, in
routing: A novel application of collaborative path-finding to smart parking 2005, as an Assistant Professor, and then became
systems,’’ in Proc. IEEE 16th Conf. Bus. Infom., Jul. 2014, pp. 119–126. a Full Professor in 2012. His research interests
[22] I. F. Akyildiz, X. Wang, and W. Wang, ‘‘Wireless mesh networks: include multimedia communication, quality-of-
A survey,’’ Comput. Netw., vol. 47, no. 4, pp. 445–487, Mar. 2005.
service, and wireless networks. In 2010, 2011, and
[23] N. Mejri, M. Ayari, R. Langar, F. Kamoun, G. Pujolle, and L. Saidane,
2012, he received the Research Excellency Award
‘‘Cooperation versus competition towards an efficient parking assignment
solution,’’ in Proc. IEEE Int. Conf. Commun., Sydney, NSW, Australia, of National Changhua University of Education. In 2012, he also received
Jun. 2014, pp. 2915–2920. the Outstanding Faculty Research Award of National Changhua University
[24] D. J. Bonde, R. S. Shende, K. S. Gaikwad, A. S. Kedari, and A. U. Bhokre, of Education. He served or serves as an Editor and a Guest Editor of several
‘‘Automated car parking system commanded by Android application,’’ in technical journals. He also served or serves on several symposium chairs and
Proc. Int. Conf. Comput. Commun. Inform. (ICCCI), Coimbatore, India, technical program committees for IEEE and other international conferences.
Jan. 2014, pp. 1–4.

VOLUME 3, 2015 1591