Content uploaded by Yuehui Zhang
Author content
All content in this area was uploaded by Yuehui Zhang on Oct 07, 2022
Content may be subject to copyright.
A Testbed for Intelligent Control of Traffic Lights
at Pedestrian Crossings on a Road
Yanshan Tian, Xu Ma
School of Mathematics
and Computer Science
Ningxia Normal University
Guyuan, China
Email: albberk@163.com
maxu20046@hotmail.com
Chu Luo
School of Computing and Information Systems
The University of Melbourne
Australia
Email: chu.luo@unimelb.edu.au
Yuehui Zhang
School of Mathematical Sciences
Shanghai Jiao Tong University
Shanghai, China
Email: zyh@sjtu.edu.cn
Abstract—Suitable testbeds can facilitate and accelerate the
development of intelligent traffic systems. In this paper we pro-
pose a testbed for testers to validate their intelligent traffic light
control programs at pedestrian crossings on a road. Our testbed
involves 6 elements: the road, pedestrian crossings, vehicles,
pedestrians, traffic light control programs, and the simulator.
Testers can input multiple parameters to simulate different road
conditions. Given the traffic light controls programs, our testbed
conducts the simulation and provides visualised performance
measures including the waiting time of vehicles and pedestrians.
Also, to reflect the the lifetime cost of vehicle brake systems, our
testbed measures how many times vehicles stop. We conducted
a case study with two traffic light control programs. The results
show that the design and implementation of our testbed is feasible
and efficient in practical simulations.
Keywords—Intelligent Control; Intelligent Traffic System; Soft-
ware Testing; Traffic Light Scheduling; Smart City
I. INTRODUCTION
Intelligent traffic light control technology is an essential part
in smart city which aims to efficiently manage urban resources
based on intelligent algorithms and data collected from the
city. Intelligent traffic light control can accelerate traffic and
reduce transportation costs. Thus, researchers have proposed
various related traffic light control techniques (e.g., [1] and
[2]). However, for software validation, these techniques all
are based on customised and specialised testing environments.
None of existing software tools can simulate traffic lights of
pedestrian crossings on a single road (i.e., traffic lights that
are not located at junctions). For instance, Figure 1 shows a
set of traffic lights of pedestrian crossings on a single road.
In the development of context-aware systems, suitable test-
ing or simulation tools can significantly reduce the time and
cost of software validation [3]. Simulated context can help
testers avoid expensive tests in real-world scenarios. As a
results, many software testing tools are designed for various
context-aware systems. Especially, a large number of tools
focus on mobile context-aware applications for the Android
platform, such as [4] and [5]. Based on these tools, Android
developers can exercise their programs and conduct testing
by simulating the context. Both functional and non-functional
properties can be checked by these tools. However, few testing
tools are proposed for intelligent traffic systems. As intelligent
traffic systems can be complex [6], quality assurance of such
systems is a challenging and time-consuming task. Due to
safety considerations, testers cannot test their programs in real
world. Moreover, in multiple iterations of development, testers
can hardly conduct regression testing (i.e., repeated testing
on a new version of software) without suitable testing tools.
Hence, it is necessary to design such a tool for researchers
and developers to conduct low-cost and efficient simulations.
To alleviate this problem, in this paper we propose a
testbed for testers to validate their intelligent traffic light
control programs at pedestrian crossings on a road. Our testbed
involves 6 elements: the road, pedestrian crossings, vehicles,
pedestrians, traffic light control programs, and the simulator.
Testers can input multiple parameters to simulate different
road conditions, including different road traces, numbers of
vehicles/pedestrians, speeds of vehicles, and positions of traffic
lights. Given the traffic light controls programs, our testbed
conducts the simulation and provides visualised performance
measures including the average and maximal waiting time of
vehicles and pedestrians. Also, to reflect the the lifetime cost of
vehicle brake systems, our testbed measures how many times
vehicles stop. To evaluate our testbed, we conducted a case
study with two traffic light control programs. The results show
that the design and implementation of our testbed is feasible
and efficient in practical simulations.
Developers of intelligent traffic light control programs can
use our testbed to validate their programs in different road
conditions. Also, testbed designers can rely on our testbed to
build new testing tools with more advanced features.
II. RE LATE D WORK
A. Traffic Light Control for Pedestrian Crossings
Intelligent traffic light control technology is an essential
part in smart city which aims to efficiently manage urban
resources based on intelligent algorithms and data collected
from the city. Intelligent traffic light control can accelerate
traffic and reduce transportation costs. Also, intelligent traffic
light control plays a critical role in the safety of citizens,
especially at pedestrian crossings.
Fig. 1. Traffic lights at a pedestrian crossing on a road (not located at
junctions).
Hence, researchers have proposed various related tech-
niques. For instance, Zhang et al. [1] present a traffic light
scheduling strategy considering both pedestrians and vehicles.
They build a novel mathematical model to represent pedes-
trians and vehicles at junction. They solve the traffic signal
scheduling problem by transforming it to a mixed integer
quadratic programming problem (which can be solved by
several existing tools).
The genetic algorithm can also facilitate the traffic light
control system at pedestrian crossings [2] and other scenarios
[7]. Based on the genetic algorithm, [2] is a technique provid-
ing intelligent green interval responses according to dynamic
traffic load, outperforming the conventional traffic controllers.
In addition, some traffic light control systems at pedestrian
crossings rely on fuzzy logic (logic system based on real
numbers between 0 and 1). For example, [8] is an approach
applying fuzzy logic to control traffic lights at pedestrian
crossings. The controller can simulate the decision mech-
anism of an experienced crossing guard. The experimental
results show that, with very simple parameter settings, the
fuzzy logic controller outperformed the conventional demand-
actuated control.
Beyond a single controller, some traffic light control tech-
niques employ multi-agent systems which consist of a group
of computer software entities. For example, [9] and [10] use
different algorithms and integrate many agents into a system
to form the control strategy. The decision process considers
diverse aspects including fluency, cost and safety.
In addition, [11] is a traffic light control technique using
FPGA as hardware. With road sensors, it can adjust the
traffic lights autonomously depending on traffic conditions.
The design is able to operate for 24 hours and 3 junctions
(12 roads) in real time.
However, these techniques all rely on customised and spe-
cialised testing environments. None of existing software tools
aim at traffic lights of pedestrian crossings on a single road
(i.e., traffic lights that are not located at junctions). It is
necessary to design such a tool for researchers and developers
to conduct low-cost and efficient simulations.
B. Testing Intelligent and Context-Aware Systems
Intelligent traffic light control is a kind of intelligent and
context-aware systems in smart cities. Many software testing
tools are designed for other kinds of context-aware systems.
For example, an early work [3] aims to provide a testing tool
for context-aware middleware-centric programs. It can abstract
data flow associations to model the structure of context-
awareness. On an RFID-based location sensing application,
experimental results show that this approach is effective and
applicable. Similarly, [12] is a testing technique that automat-
ically generates suitable test cases by analysing application
description documents. The analysis of semantic annotations
can help this technique exercise the application more carefully.
Also, to ensure the efficiency of testing, this technique can
detect and remove redundant test cases. Also, [4] is a testing
tool for Android context-aware applications. It allows testers
to conduct tests by replaying recorded data such as GPS traces.
However, a limitation of it is that it can only replay the data
recorded by itself, rather than data from other sources (e.g.,
online databases). As a result, some recent tools allow testers
to conduct tests by replaying data from other sources. For
example, KnowMe [13], [14] and TestAWARE [5] all accept
datasets exported from the data collection middleware AWARE
[15]. Moreover, testers can use simulated data and real-time
data together with recorded data in [14] and TestAWARE [5].
This data integration increases the coverage of application
running states during tests. TestAWARE [5] can also measure
the real-time performance for some time-critical applications
(e.g., real-time sensing systems [16], [17], [18]).
Compared to data-driven approaches, [19] is a model-based
approach for testing mobile context-aware applications. It can
avoid the loss of coverage caused by imbalanced distributions
of different conditions in testing data.
Since these testing approaches can simplify the development
of some kinds of context-aware systems, it is a meaningful task
to develop similar software testing tools aiming at traffic lights
of pedestrian crossings on a single road. Thus, in this paper
we present a testbed for such a task.
III. TES TB ED DESIGN
To test the control programs of traffic lights at pedestrian
crossings on a road, our testbed involves 6 elements: the road,
pedestrian crossings, vehicles, pedestrians, traffic light control
programs, and the simulator. Figure 2 shows the architecture
of our testbed. And Figure 3 shows the meta model of our
testbed.
A. Modelling the Road
To model the road in a realistic way, our testbed takes as
input any real-world GPS trace file in gpx format. These files
Fig. 2. Testbed architecture.
Fig. 3. Meta model of the testbed.
can easily be found at map sites such as OpenStreetMap [20].
Since GPS traces are normally sampled per second, given the
relative speed compared to the sampler (note that the speed
of samplers may vary due to the dynamic road condition),
a certain number of seconds can determine the locus on the
road.
B. Modelling Pedestrian Crossings
With a GPS trace, testers can set a number of GPS co-
ordinates or the timings of the corresponding collected GPS
coordinates on this trace to represent pedestrian crossings
on the road. Accordingly, each pedestrian crossing will have
a traffic light which lets the movements of pedestrians and
vehicles be two mutually exclusive events.
C. Modelling Vehicle Traffic on the Road
To model vehicle traffic, testers have to input 3 parameters:
the number of vehicles, the maximal and minimal relative
speed of vehicles. The baseline of the speed is the speed of
the sampler that generated the GPS trace. Using the maximal
and minimal relative speed, our testbed generates the expected
number of vehicles having a random speed in the middle part
of a normal/Gaussian distribution which is commonly used.
The density of a vehicle with speed xcan be given by:
D(x) = 1
σ√2πe
−(x−µ)2
2σ2, x < max, x > min (1)
where max and min are the maximal and minimal relative
speed of vehicles, µis the baseline speed, and σ2is the
variance (default σvalue is 0.4 so that the vehicles have a
speed generally close to the baseline. Testers can change the
value, and even the distribution function, depending on their
need). Note that the density (the sum is not necessarily 1) is
not the probability (the sum is 1) due to the range of xdoes
not cover all the real numbers.
Half of the vehicles start from the beginning of the road,
and others start from the end of the road. Vehicles move to the
other end of the road at their own speed, respectively. Each
vehicle has a timer for our testbed to count how much time is
spent on waiting for traffic lights. Also, each vehicle counts
how many times it stops to wait.
D. Modelling Pedestrians
Since pedestrians only cross the road, rather than walking
along, our testbed randomly generate pedestrians at each
crossing in the whole period of simulation. The upper bound
of the simulation duration depends on the slowest vehicle:
Duration ≤min(roadLength/x) +
n
X
i=1
Ri(2)
where roadLength is the length of the road, xis the speed of
a vehicle, Riis the red light duration of the number itraffic
light. This inequality means that the simulation duration is not
longer than the time for the slowest vehicle to meet all full
red lights at each crossing.
However, considering that intelligent control programs may
dynamically determine the traffic lights, Riis not always
known before simulation. Hence, our testbed keeps gener-
ating new pedestrians randomly at each moment until the
last vehicle (not necessarily the slowest vehicle) or the last
pedestrian, whichever later, finishes the trip. Each moment in
the simulation has the same likelihood of generating some new
pedestrians. The average number of pedestrians to generate per
second is defined by testers.
Similar to vehicles, each pedestrian has a timer to count how
much is spent on waiting for traffic lights. Since pedestrians
mostly walk at similar speed, each pedestrian has a default
speed (20s to pass). Testers can change the pedestrian speed
depending on need.
E. Traffic Light Control
Traffic light control programs are provided by testers.
Whether the programs are intelligent or not, they should output
the actions to adjust or maintain the states of traffic lights.
Intelligent control programs should consider nearby vehicles,
pedestrians, the current state of the controlled traffic light, and
even the states of nearby traffic lights. Mathematically, based
on such dependencies, intelligent traffic light control programs
can be modelled as functions:
f(Lt−n...t, Vt−n...t , Pt−n...t) = Lt+1 (3)
where Ltis the states of all the traffic lights at moment t;Vt
vehicles; Ptpedestrians; nis the historical states affecting the
decision-making process of programs.
TABLE I
PERFORMANCE MEASURES FOR VEHICLES AND PEDESTRIANS
Vehicles Pedestrians
Waiting Time Mean, Max Mean, Max
Stops Mean, Max N/A
F. Simulation Mechanism
For successful simulation, the simulator of our testbed has
two tasks:
1) To call and apply the traffic light control. When calling
the traffic light control, the simulator has to provide
sufficient information (e.g., the number and positions
of nearby vehicles and pedestrians) required by the
program.
2) To determine new states of vehicles and pedestrians.
When the simulation is ongoing, vehicles and pedes-
trians continue moving to their destinations unless the
traffic light is red. If they stop to wait for the traffic light,
the simulator has to count the waiting time for each of
them.
The two tasks are executed simultaneously. The states
of traffic lights, vehicles and pedestrians are updated every
second. The simulator finally uses the recorded waiting time to
evaluate the effectiveness of each traffic light control program.
Specifically, as shown in Table 1, our testbed has measures for
vehicles and pedestrians in two aspects. By measuring how
many times vehicles stop to wait for traffic lights, our testbed
can reflect the lifetime cost of vehicle brake systems. Finally,
our testbed outputs the results via bar charts.
G. Implementation
We implemented our testbed as a Python 3.6 package. This
can allow testers to test various intelligent algorithms and
machine learning techniques in Python packages, such as the
popular deep learning library Tensorflow [21]. The GPS trace
file (gpx) is parsed by gpxpy package. The visualisation of test
results is based on Matplotlib package.
IV. CAS E STU DY
To show the effectiveness of our testbed, we ran a case
study on a PC with a simulated road, traffic lights, vehicles,
pedestrians and two control policies.
The settings of road conditions are the following:
1) The road was constructed from a GPS trace file on
OpenStreetMap [20], having 1804 seconds of a single
trip.
2) We generated 1000 vehicles (i.e., 500 for each direction)
with minimal speed 0.5X and maximal 1.5X multiple of
the original trace speed.
3) 5 traffic lights were randomly created.
4) The pedestrians at each traffic light were randomly
generated with the likelihood of 0.001 per second. Each
of them has the same speed and requires 20s to pass.
If the current green light has less than 20s, they will
choose to wait for the next green light.
Fig. 4. Results from our testbed for the fixed periodic schedule. The left
vertical axis is for waiting time. The right vertical axis is for stops.
The two tested control policies are:
1) fixed periodic schedule: the state of each traffic light is
changed per 60s.
2) road-aware intelligent control: if there are no pedestrians
at a crossing and there are incoming vehicles that will
arrive within 10s, it gives 10s green light unless the
current state is green with longer time. If the conditions
are not met (e.g., the pedestrians and vehicles all at a
crossing), it follows the fixed periodic schedule.
The random seed was set to 1 for the reproduction of the road
conditions for both control policies.
A. Results
The both tests were completed in one minute. In the tests,
the number of generated pedestrians was 14.
Figure 4 shows the test results from our testbed for the
fixed periodic schedule. For vehicles, the mean waiting time
is 91.998s and the maximal waiting time is 293.637s. The
mean stops are 3.195; maximal 5. For pedestrians, the mean
waiting time is 24.571s and the maximal waiting time is 66s.
Figure 5 shows the test results from our testbed for the road-
aware intelligent control. For vehicles, the mean waiting time
is 1.664s and the maximal waiting time is 29.134s. The mean
stops are 0.175; maximal 2. For pedestrians, the mean waiting
time is 7.071s and the maximal waiting time is 49s.
Through the results visualised by our testbed, we can
conduct efficient comparison and draw conclusions. The road-
aware intelligent control is significantly better than fixed
periodic schedule for both vehicles and pedestrians in this case
of road conditions.
Fig. 5. Results from our testbed for the road-aware intelligent control. The
left vertical axis is for waiting time. The right vertical axis is for stops.
V. DISCUSSION
A. Implications
We show that the tests of intelligent traffic light control at
pedestrian crossings on a road can be conducted efficiently
using our testbed. Since intelligent traffic light control plays a
crucial role in the safety and happiness of citizens at pedestrian
crossings, our testbed can be used or reconfigured to validate
and evaluate the intelligent traffic light algorithms. Also, smart
cities aim to achieve higher energy-efficiency for traffic. Based
on the measures of vehicle waiting time and stops, our testbed
can be used to quantify the energy-efficiency of different traffic
light control programs.
In addition, the road conditions are very different among
cities and even city areas. The time of a day is also related to
the state of traffic. An intelligent traffic light control program
may have different performance under different road condi-
tions. To consider the different road conditions, our testbed
can take different parameters (e.g., number of vehicles, their
speed and the likelihood of the generation of pedestrians) to
model the volume of traffic, for both vehicles and pedestrians.
However, determining these parameters is be a difficult task
for human, mining historical data or using other smart city
techniques (e.g., smart-city-based video surveillance [22]) can
help human to find a suitable set of parameters.
B. Limitation and Future Work
Our testbed contains limitations. First, it does not take
cyclists into account. The cyclists are also a part of road
traffic, requiring attention from pedestrians. However, the trips
of cyclists do not affect the energy-efficiency of our society,
compared to vehicles. Some research has focused on smart
traffic systems for cyclists, such as smart traffic scheduling
using smartphones of cyclists [23]. Hence, future work may
regard cyclists as a unique element in the testbed design.
Second, our testbed does not consider the dynamic driving
strategies of vehicles. For example, in vehicular networks [24],
vehicles may communicate with each other and adjust their
speed to pass a crossing together during a green light. Future
cyclists can also do the same thing using device-to-device
communication [25]. Future work can investigate the testbed
design for the dynamic strategies of vehicles and cyclists.
Lastly, our testbed does not measure the computational cost
of intelligent traffic light control programs. In practice, it is
important to ensure the fast response of programs in real-time
computing tasks. Future work can provide this feature.
VI. CONCLUSION
In this paper we propose a testbed for testers to validate
their intelligent traffic light control programs at pedestrian
crossings on a road. Our testbed involves 6 elements: the
road, pedestrian crossings, vehicles, pedestrians, traffic light
control programs, and the simulator. Testers can input multiple
parameters to simulate different road conditions. Given the
traffic light controls programs, our testbed conducts the simu-
lation and provides visualised performance measures including
the waiting time of vehicles and pedestrians. Also, to reflect
the the lifetime cost of vehicle brake systems, our testbed
measures how many times vehicles stop. We conducted a case
study with two traffic light control programs. The results show
that the design and implementation of our testbed is feasible
and efficient in practical simulations.
ACKNOWLEDGMENT
This work is supported by NSFC grant 11671258,
11771280, National Science Foundation of Shanghai Munic-
ipal (17ZR1415400), National Natural Science Foundation
of China under grant No. 11361046, Ningxia High Educa-
tion research fund with grant No. NGY2017180, undergrad-
uate teaching project of Ningxia High Education with grant
No.NXJG2016060, and the fund of Ningxia High Educa-
tion Construction of First-Class Disciplines (Education) with
grant No.NXYLXK2017B11. and Key Research and Develop-
ment (Science and Technology Support Program) Program of
Ningxia Province No.2018BEE03025, 2018BEE03026.
REFERENCES
[1] Y. Zhang, R. Su, K. Gao, and Y. Zhang, “Traffic light scheduling
for pedestrians and vehicles,” in Control Technology and Applications
(CCTA), 2017 IEEE Conference on. IEEE, 2017, pp. 1593–1598.
[2] A. M. Turky, M. S. Ahmad, and M. Z. M. Yusoff, “The use of genetic
algorithm for traffic light and pedestrian crossing control,” International
Journal of Computer Science and Network Security, vol. 9, no. 2, pp.
88–96, 2009.
[3] H. Lu, W. Chan, and T. Tse, “Testing context-aware middleware-centric
programs: a data flow approach and an rfid-based experimentation,” in
Proceedings of the 14th ACM SIGSOFT international symposium on
Foundations of software engineering. ACM, 2006, pp. 242–252.
[4] Z. Qin, Y. Tang, E. Novak, and Q. Li, “Mobiplay: A remote execution
based record-and-replay tool for mobile applications,” in Proceedings
of the 38th International Conference on Software Engineering. ACM,
2016, pp. 571–582.
[5] C. Luo, M. Kuutila, S. Klakegg, D. Ferreira, H. Flores, J. Goncalves,
M. M¨
antyl¨
a, and V. Skostakos, “Testaware: A laboratory-oriented infras-
tructure for mobile context-aware applications,” Proceedings of the ACM
on Interactive, Mobile, Wearable and Ubiquitous Technologies, vol. 1,
no. 3, p. 80, 2017.
[6] R. T. van Katwijk, P. van Koningsbruggen, B. De Schutter, and
J. Hellendoorn, “A test bed for multi-agent control systems in road
traffic management,” in Applications of Agent Technology in Traffic and
Transportation. Springer, 2005, pp. 113–131.
[7] A. M. Turky, M. Ahmad, M. Yusoff, and B. T. Hammad, “Using genetic
algorithm for traffic light control system with a pedestrian crossing,” in
International Conference on Rough Sets and Knowledge Technology.
Springer, 2009, pp. 512–519.
[8] J. Niittymaki and S. Kikuchi, “Application of fuzzy logic to the con-
trol of a pedestrian crossing signal,” Transportation Research Record:
Journal of the Transportation Research Board, no. 1651, pp. 30–38,
1998.
[9] I. Kosonen, “Multi-agent fuzzy signal control based on real-time simula-
tion,” Transportation Research Part C: Emerging Technologies, vol. 11,
no. 5, pp. 389–403, 2003.
[10] M. Abdoos, N. Mozayani, and A. L. Bazzan, “Traffic light control
in non-stationary environments based on multi agent q-learning,” in
Intelligent Transportation Systems (ITSC), 2011 14th International IEEE
Conference on. IEEE, 2011, pp. 1580–1585.
[11] S. Nath, C. Pal, S. Sau, S. Mukherjee, A. Roy, A. Guchhait, and
D. Kandar, “Design of an fpga based intelligence traffic light controller
with vhdl,” in Radar, Communication and Computing (ICRCC), 2012
International Conference on. IEEE, 2012, pp. 92–97.
[12] R. Tonjes, E. S. Reetz, M. Fischer, and D. Kuemper, “Automated testing
of context-aware applications,” in Vehicular Technology Conference
(VTC Fall), 2015 IEEE 82nd. IEEE, 2015, pp. 1–5.
[13] S. Bobek, “Contextsimulator,” http://glados.kis.agh.edu.pl/doku.php?id=
pub:software:contextsimulator:start, May 2017.
[14] C. Luo, M. Kuutila, S. Klakegg, D. Ferreira, H. Flores, J. Goncalves,
V. Kostakos, and M. M¨
antyl¨
a, “How to validate mobile crowdsourcing
design? leveraging data integration in prototype testing,” in Proceedings
of the 2016 ACM International Joint Conference on Pervasive and
Ubiquitous Computing: Adjunct. ACM, 2016, pp. 1448–1453.
[15] D. Ferreira, V. Kostakos, and A. K. Dey, “Aware: mobile context
instrumentation framework,” Frontiers in ICT, vol. 2, p. 6, 2015.
[16] C. Luo, A. Fylakis, J. Partala, S. Klakegg, J. Goncalves, K. Liang,
T. Sepp¨
anen, and V. Kostakos, “A data hiding approach for sensitive
smartphone data,” in Proceedings of the 2016 ACM International Joint
Conference on Pervasive and Ubiquitous Computing. ACM, 2016, pp.
557–568.
[17] C. Luo, H. Koski, M. Korhonen, J. Goncalves, T. Anagnostopoulos,
S. Konomi, S. Klakegg, and V. Kostakos, “Rapid clock synchronisa-
tion for ubiquitous sensing services involving multiple smartphones,”
in Proceedings of the 2017 ACM International Joint Conference on
Pervasive and Ubiquitous Computing and Proceedings of the 2017 ACM
International Symposium on Wearable Computers. ACM, 2017, pp.
476–481.
[18] S. Klakegg, C. Luo, J. Goncalves, S. Hosio, and V. Kostakos, “In-
strumenting smartphones with portable nirs,” in Proceedings of the
2016 ACM International Joint Conference on Pervasive and Ubiquitous
Computing: Adjunct. ACM, 2016, pp. 618–623.
[19] T. Griebe and V. Gruhn, “A model-based approach to test automation for
context-aware mobile applications,” in Proceedings of the 29th Annual
ACM Symposium on Applied Computing. ACM, 2014, pp. 420–427.
[20] OpenStreetMap, “Public gps traces,” http://www.openstreetmap.org/
traces, Feb. 2018.
[21] Google, “Building mobile apps with tensorflow,” https://www.tensorflow.
org/mobile/, Feb. 2018.
[22] C. Luo, “Video summarization for object tracking in the internet of
things,” in Next Generation Mobile Apps, Services and Technologies
(NGMAST), 2014 Eighth International Conference on. IEEE, 2014,
pp. 288–293.
[23] T. Anagnostopoulos, D. Ferreira, A. Samodelkin, M. Ahmed, and
V. Kostakos, “Cyclist-aware traffic lights through distributed smartphone
sensing,” Pervasive and Mobile Computing, vol. 31, pp. 22–36, 2016.
[24] H. Moustafa and Y. Zhang, Vehicular networks: techniques, standards,
and applications. Auerbach publications, 2009.
[25] H. Flores, R. Sharma, D. Ferreira, C. Luo, V. Kostakos, S. Tarkoma,
P. Hui, and Y. Li, “Social-aware device-to-device communication: a
contribution for edge and fog computing?” in Proceedings of the
2016 ACM International Joint Conference on Pervasive and Ubiquitous
Computing: Adjunct. ACM, 2016, pp. 1466–1471.