Content uploaded by Roger Immich
Author content
All content in this area was uploaded by Roger Immich on Aug 07, 2018
Content may be subject to copyright.
Shielding video streaming against packet losses over VANETs
Roger Immich
1
•Eduardo Cerqueira
2
•Marilia Curado
1
Published online: 13 November 2015
ÓSpringer Science+Business Media New York 2015
Abstract Vehicular ad-hoc networks (VANETs) are
being widely adopted in the last few years. This type of
network enables the utilization of a large diversity of dis-
tributed applications, such as road and traffic alerts,
autonomous driving capabilities and video distribution.
Video applications can be considered one of the most
demanding services because it needs a steady and contin-
uous flow of information. This presents a set of challenges
to VANETs considering their scarce network resources due
to the vehicle movement and time-varying wireless chan-
nels. Considering the above mentioned issues, an adaptive
quality of experience (QoE)-driven mechanism is needed
to provide live transmission capabilities to video-equipped
vehicles. This mechanism has to overcome the challenges
to grant a high-quality video transmission without adding
any unnecessary network overhead. To this end, a forward
error correction (FEC) technique can be adapted to enhance
the video distribution, leading to higher QoE for end users.
The proposed self-adaptive FEC-based mechanism
(SHIELD) uses several video characteristics and specific
VANETs details to safeguard real-time video streams
against packet losses. One of the main contributions of this
work is the combined used of network density, signal-to-
noise ratio, packet loss rate, and the vehicle’s position. This
allows SHIELD to better protect the video sequences and
enhance the QoE. In doing that, we are able to improve the
user experience, while saving network resources. The
advantages and drawbacks of the proposed mechanism are
demonstrated through extensive experiments and assessed
with QoE metrics, proving that it outperforms both adap-
tive and non-adaptive mechanisms.
Keywords VANETs Forward error correction (FEC)
Unequal error protection (UEP) Fuzzy logic Quality of
experience (QoE)
1 Introduction
The vehicular ad-hoc network (VANET) is considered the
core component of intelligent transportation systems (ITS),
providing support to many applications, including video
services. This type of service is gaining popularity on a
daily basis, being currently on high demand [1,2] due to
the popularization of better network and video devices. The
adoption of video services can be used as means to provide
users with both information and entertainment content.
The endorsement of video-based services can be bene-
ficial to a broad range of situations, such as road safety,
driver awareness, traffic status, and infotainment applica-
tions. Besides the users’ experience, the video quality is
also important to allow a better assessment of each situa-
tion. For example, it can give police officers, paramedics,
and fire fighters an accurate representation of the scene
they will attend, thereby reducing the response time.
Beyond the traffic related services, a sport venue or a
festival could broadcast a live feed to incoming fans caught
in a traffic jam. These are only simple examples of a
&Roger Immich
immich@dei.uc.pt
Eduardo Cerqueira
cerqueira@ufpa.br
Marilia Curado
marilia@dei.uc.pt
1
Department of Informatics Engineering, University of
Coimbra, Coimbra, Portugal
2
Faculty of Computer Engineering, Federal University of Para,
Belem, BR, Brazil
123
Wireless Netw (2016) 22:2563–2577
DOI 10.1007/s11276-015-1112-z
limitless number of alternatives to make available rich-
format services. These services, however, have to deal with
the unreliable wireless connection of the VANETs, which
are highly dynamic in nature and strongly prone to packet
loss [3,4]. Because of that, it is imperative to strengthen
the video transmissions against losses [5,6]. This calls for
an adaptive mechanism to enhance the video delivery to
provide higher Quality of Experience (QoE).
QoE is a set of methods to assess the overall customer’s
experience level of satisfaction regarding a service. This
method is related to, but differs, from the well-studied
Quality of Service (QoS). In VANETs there is still a lack of
adaptive QoE-driven mechanisms to better support live
video transmissions [7–10]. This can be attributed to the
challenging combination of the VANETs’ dynamic topol-
ogy and the stringent video requirements. In order to sur-
pass these adversities, a good mechanism has to take into
consideration several aspects of the intrinsic network
characteristics and video details, being able to correctly
identify and protect the most QoE-sensitive data.
Several techniques have been proposed to tackle the
VANETs challenges in the last few years. Some of them are
trying to solve these issues throughout adaptive routing
protocols [11–15]. The results show that a reliable routing
protocol has a major influence on improving the video
quality. This improvement, however, is restricted to a
specific level. After this level, to increase or even sustain the
video quality it is crucial to resort to some amount of
redundant data, which allows reconstructing the original data
set in case of packet losses. A known approach to supply this
redundancy is using Forward Error Correction (FEC) tech-
niques. These techniques have been adopted and produced
favourable outcomes by enhancing the video quality in live
transmissions [16,17]. However, due to the video require-
ment of a timely delivery of a considerable amount of data
[18], along with the shared wireless channel resources, a self-
adaptive FEC-based mechanism is advisable. This mecha-
nism needs to have the capability to operate under unfore-
seen conditions in order to increase the human perception,
while reducing the network overhead.
In order to tackle the above-mentioned issues, this
article proposes a self-adaptive FEC-based proactive error
recovery mechanism to shield video transmissions over
VANETs (SHIELD). One problem frequently found in
FEC-based mechanisms is absence of QoE-related details
to compute the required amount of redundancy. For this
reason, SHIELD is also a QoE-driven mechanism. This
means that meaningful video aspects related to the human
point-of-view, are not neglected, which leads to the addi-
tion of a very specific amount of redundancy.
Another important feature of the proposed mechanism is
the use of Unequal Error Protection (UEP). Not all video
packets have the same importance to ensure the final video
quality [19,20]. To improve in these issues, SHIELD
adopts a Hierarchical Fuzzy System (HFS) [21]. HFS
allows adding an accurate amount of video redundancy
specifically to the more QoE-sensitive data. This increases
the video quality according to the human perception while
cutting down on the network overhead.
The SHIELD mechanism was evaluated using real video
sequences and actual maps’ clippings with the aid ofobjective
QoE metrics. The remainder of this article is organised as
described next. Sect. 2features the related work. Sect. 3
describes the SHIELD mechanism and Sect. 4its assessment.
Conclusions and future work are presented in Sect. 5.
2 Related work
In recent years, several techniques have been proposed to
increase the quality of video transmission over VANETs.
Some of these proposals rely on routing protocol adapta-
tions, e.g. the QoE-based routing protocol for video
streaming over VANETs (QOV) [13]. In QOV, the per-
ceptual quality of the videos is assessed in real-time, at the
receivers, using the Pseudo-Subjective Quality Assessment
(PSQA) [22] metric. After that, the results are announced
to the neighbours throughout Hello packets. This allows
the routing protocol to choose the best paths to deliver the
video sequences. Nevertheless, VANETs are very dynamic
networks and because of that, the proposed mechanism
would have to update very quickly the PSQA result
announcement, overloading the network with Hello
packets. Another weakness of this proposal is that it does
not include any type of error correction (EC). As afore-
mentioned, the video quality can be maintained only up to
a certain level without using EC, however, if the network
has a high packet loss rate the quality will decrease.
Another proposal is an adaptive multi-objective Medium
Access Control (MAC) retransmission limit strategy [23].
At the Road Side Units (RSUs), channel statistics and
packet transmission rate are used as input to the opti-
mization framework in order to tune the MAC retrans-
mission limit. This optimization improves the performance
of video transmission, leading to better video quality.
However, it aims to only minimize the playback freezes
and reduce the start-up delay. These are important char-
acteristics, however, QoE metrics should be used to assess
the image quality. This evaluation would provide a more
comprehensive assessment of the proposed mechanism.
Additionally, the authors only took into account the use of
RSUs and two-hop communications. It is known that the
major advantages of VANETs come from the communi-
cation directly between the vehicles, without the need for a
fixed infrastructure. This severely restricts the application
of the mechanism.
2564 Wireless Netw (2016) 22:2563–2577
123
In addition to these mechanisms, several FEC-based
methods have also been proposed to enhance the quality of
videos in transmissions over VANETs. The Hybrid Video
Dissemination Protocol (HIVE) [24] uses a multi-layer
strategy to improve the video quality. The HIVE multi-
layer strategy is based on the joint use of traffic congestion
control scheme, node selection method, and application
layer erasure coding technique. This allows higher packet
delivery ratio, while keeping latency and packet collisions
low. The results show improvement in the Peak signal-to-
noise ratio (PSNR) assessment, leading the authors to claim
that they improved the QoE for end-users. However,
relying in only one metric is not enough to prove that,
especially considering that the PSNR results do not cor-
relate well with the human vision system [25]. Another
issue is the lack of video characteristics assessment. It is
known that these video details have a considerable impact
on how resilient a video sequence is against packet loss.
The Blind XOR (BXOR) scheme [26] adopts an adaptive
low-overhead XOR technique to enhance the video quality.
This mechanism works by blindly setting packets to be
retransmitted, relying on the conditional reception probability
(CRP). This means that, if there is a probability of not
receiving a set of packets, they are tagged to be retransmitted,
even if they had not been lost yet. The estimation of the CRP is
performed on the server side without feedback from the cli-
ents. A drawback of this mechanism is that it heavily relies on
the CRP estimation, which may not be accurate. Furthermore,
this mechanism also does not take into consideration the video
characteristics. As previously mentioned, this detail can have
a significant impact on the video quality, especially on
determining a precise amount of redundancy.
Another mechanism to improve the video quality over
VANETs compares the efficiency of Random Linear Coding
(RLC) and XOR-based coding [27]. The benchmark results
show that both erasure codes are able to improve the video
quality by increasing the number of successfully received
packets over error-prone networks. The results also show
that XOR-based coding outperforms the RLC scheme. In
addition, the proposed mechanism finds the optimal packet
block size, which allows adding a more precise amount of
redundancy. However, important features are not consid-
ered, namely the network status and the video characteris-
tics. These details play a critical role in the optimization of
the amount of redundancy required to provide both good
video quality and low network overhead.
3 QoE-driven video transmission
On account of the previously mentioned challenges, this
work presents and assesses the self-adaptive FEC-based
SHIELD mechanism. The importance of this proposal
relies on the shortage of QoE-driven mechanisms that are
able to combine video characteristics, such as the motion
activity, with particular VANETs features. Our mechanism
is able to offer videos with higher QoE while, at the same
time, downsizing the network overhead footprint. This
work improves on our previous mechanism [28]. Key
enhancements are disclosed and discussed below.
Additionally, in this work the vehicle-to-vehicle (V2V)
communication characteristics are explored to better adjust
the proposed mechanism to the actual network conditions.
Even though a VANET environment enables roadside
infrastructure, the V2V communication was chosen
because it is unlikely that such infrastructure will cover all
the highways and cities in the near future. Consequently, if
the infrastructure is available it can be used, however, the
optimizations will only be performed on the communica-
tion between the vehicles.
3.1 SHIELD overview
Figure 1depicts an overview of the proposed mecha-
nism. The first step, is to assess the network conditions (1).
In order to do that, different parameters are evaluated in a
combined way, namely the network density, SNR, and
PLR, as well as the node’s position. To calculate the
density, first the network area is found through an
approximate hull algorithm. After that, the total number of
1-hop nodes is divided by the area, which gives the net-
work density. All these parameters are necessary because
none of them by itself is accurate enough to characterize
the quality on the network links [29,30]. The combination
of them, however, can provide a very good estimation of
the network conditions. Thereafter, using cross-layer
techniques, important details about the video characteris-
tics are collected (2). In the video-aware procedure of the
mechanism several details are analysed, such as the image
resolution, frame type and size, motion vectors, and mac-
roblock configuration. At the end, all the gathered data are
fed to the fuzzy inference engine, which will compute a
specific amount of redundancy (3).
Provided that the network conditions are not the same at
all intermediate nodes, this parameter has to be reassessed
at each hop (4). On the other hand, the video characteristics
do not change during the transmission. Because of that,
they are embedded in each packet header by the server
node. This eliminates the need for processor intensive tasks
(e.g. deep packet inspection) on each and every packet. The
IPv6 optional hop-by-hop header was chosen to store this
information [31]. This means that it is always ready to use
whenever needed (5,6). Owing to this, the task of adjusting
the redundancy amount on each hop is facilitated. The
Wireless Netw (2016) 22:2563–2577 2565
123
result is a higher video quality, and consequently, superior
QoE is perceived by the end-users (7).
3.2 Towards the design of SHIELD
This section describes the manifold procedure and modules
that the SHIELD mechanism is composed of. Primarily, to
enable the SHIELD real-time capabilities, a knowledge
database is needed. This database is created using a hier-
archical clustering technique [5] to store video details,
which includes several video characteristics and their
impact on the QoE. A comprehensive description of this
process can be found in [32].
Another important feature to enable the SHIELD real-
time capabilities is the use of Fuzzy Logic (FL). This
allows building a dynamic and comprehensive scheme,
which takes into consideration several network and video
characteristics, and still manages to perform in real-time.
Nevertheless, in conventional FL systems the rules grow
exponentially according to the number of variables.
Because of that, it is common to have a rule-explosion
situation when handling a lot of variables, making the FL
controller very hard to implement. To address this issue,
the SHIELD mechanism was designed to use a Hierarchi-
cal Fuzzy System (HFS). In HFS, low-dimensional fuzzy
systems can be arranged in a hierarchical form, reducing
the global number of rules because the system grows
linearly.
The combined use of the knowledge database and
human expertise enables setting up the fuzzy sets, rules,
and hierarchical levels. Once this analysis is finished, the
produced data is loaded in the fuzzy inference engine,
making it possible to be performed in real-time. This is a
very important step in the mechanism because it reduces
the number of rules that will be processed in real-time,
leading to a more precise and faster mechanism.
Figure 2depicts the hierarchical levels adopted by
SHIELD. There are three layers, namely (A) Objective
function, (B) General criteria, and (C) Specific criteria. The
output of each low-level layer is used as input to the next
layer. The first layer (A) represents the amount of redun-
dancy that our mechanism will add to a specific portion of
video data. The main goal is to find the amount of redun-
dancy for a system that, given its constraints, results in less
network overhead and better QoE. The second layer
(B) encompasses the overall details that the proposed
mechanism uses to determine the redundancy amount,
namely the network details and the video characteristics.
The bottom layer (C) is responsible for handling the input
parameters of each feature used by the fuzzy logic infer-
ence system. This layer has a subdivision (C)(2), which is
performed at each network hop. All the input parameters
(C)(1) are only taken into consideration at the server node.
The design of HFS follows the same method as in
standard fuzzy logic schemes. This means that several
Fig. 1 General view of the SHIELD mechanism
Fig. 2 Hierarchical fuzzy logic structure
2566 Wireless Netw (2016) 22:2563–2577
123
fuzzy components have to be defined, such as sets, rules,
membership functions and the inference engine. The fuzzy
rules are a group of linguistic control rules, which describe
how the system works. The fuzzy sets are a collection of
elements that have some degree of membership. This dif-
fers from the classical set definition, where an element
either belongs or does not belong to a set. The membership
functions provide the degrees of truth of each element in
the fuzzy set. The inference engine is responsible for the
decision-making process, which is based on the fuzzy rules,
sets and the input linguistic parameters. This is an offline
process and needs to be executed only once. Following
this, the resulting data are loaded into the fuzzy inference
engine to be used in real-time. A detailed explanation of
this process is given below.
3.2.1 The ‘‘general network’’ criteria
The ‘‘general network’’ criteria accounts for the definition of
the network conditions. The characterization of a good or bad
channel is not an easy task and it cannot rely upon a single
metric, especially in wireless networks [29]. With this in
mind, the SHIELD mechanism uses four metrics to better
establish a network quality indicator. These metrics are
divided into two specific criteria, namely ‘‘network status’’
and ‘‘communication surroundings’’. The former is defined
by the combined assessment of the SNR and the PLR. The
latter is given by the network density and the position of the
vehicles. Each one of these metrics is described next.
The SNR is the level of the desired signal against the
level of background noise. This is a good indicator for the
physical medium, especially for spectrum sensing. While
this is true, it cannot be considered a reliable general net-
work quality indicator by itself. This steams from the fact
that a strong channel signal will not always produce a good
network connection [29]. On the other hand, a very weak
signal will yield a low quality network connection. Because
of that, to create a more holistic indicator more than one
metric has to be used. Another obvious candidate to define
the network quality is the PLR. In general, the SNR and
PLR have a negative correlation, meaning that when one
increases, the other decreases and vice versa. However,
they complement each other because the SNR takes into
consideration the physical spectrum part of the transmis-
sion and the PLR provides a point of view closer to the
application layer.
An extensive number of network simulations were car-
ried out to better characterize the impact of different PLRs
in the QoE. Video sequences tend to have a natural resi-
liency to packet loss [33], because of that, several video
sequences, with distinct features were used during the
experiments. The output of the experiment made evident
that it is possible to have a good QoE with packet loss
between 0 and 12 %. In most of the cases, an accept-
able video quality for end-users was perceived with losses
from 5 up to 23 %. However, after a threshold of 19 % the
video quality starts to decrease apace, particularly in videos
with high resolution and motion intensity. In the experi-
ments with more than 36 % of PLR the QoE reached
unbearable levels. Algorithm 1 shows only one of the many
fuzzy sets defined in the SHIELD mechanism. In this case,
it is the PLR fuzzy set, which was found through the
experiments aforementioned.
Algorithm 1: Packet loss rate fuzzy set
InputLVar* PLR = new InputLVar(”PacketLossRate”);
PLR →addTerm( TriangularTerm(”LOW ”, 0, 12));
PLR →addTerm( TriangularTerm(”MEDIUM ”, 5, 23));
PLR →addTerm( TriangularTerm(”HIGH ”, 19, 100));
engine.addInputLVar(PLR);
Another component of the ‘‘general network’’ criteria is
the ‘‘communication surroundings’’. As aforementioned, it
uses the network density and the position of the vehicles to
provide more information about the network in which the
video sequences are being transmitted. These parameters
are updated at each beacon exchange in the routing pro-
tocol. The network density is given by the number of
nodes, in our case vehicles, divided by the network area. It
is important to notice that VANETs are very dynamic
networks with a decentralized structure, proving to be a
challenge the estimation of the network surface area. To
address this issue, the proposed mechanism uses an
approximate convex hull algorithm.
A convex hull algorithm is able to find the smallest
boundary polygon containing all the points inside of it,
using only non-intersecting segments, as showed in Fig. 3a.
There are several algorithms to find the convex hull of a
given set of points. In our previous work [28] the Quic-
kHull [34] method was used. It uses a divide-and-conquer
algorithm with average complexity of Oðnlog nÞand at the
worst case it could take Oðn2Þ. However, the proposed
mechanism does not need a high precision value for the
area, instead, a good approximation is sufficient to provide
very good results. Because of this, and to improve the
general performance, we use the BFP [35] approximation
convex hull algorithm as showed in Fig. 3b.
The BFP algorithm, which runs in OðnÞtime, replaces
the sort operation by dividing the plane in vertical strips. In
each strip, the minimum and maximum points are found
and added to the boundary. This algorithm is an approxi-
mation because a non-extreme point, in a given strip, can
be discarded even if it is on the convex hull boundary.
Nevertheless, the point will not be far from the convex hull,
resulting in a good approximation of the actual convex
hull.
Wireless Netw (2016) 22:2563–2577 2567
123
Figure 4shows the comparison between the number of
nodes and the resulting number of operations in both
QuickHull and BFP algorithm. On average, the QuickHull
algorithm has fairly good performance, it can degrade
however, up to exponential in the worst case. On the
contrary, the BFP algorithm has a steady linear perfor-
mance, providing results more quickly even with small
number of nodes.
There is a multitude of advantages to perform fewer
operations. First of all, due to the time-sensitive video data
it is important that the client node receives the information
as soon as possible, thus performing fewer operations
allows dispatching the video quickly. In addition, because
of the fast time-varying network conditions, the faster this
information is made available the more accurate it is. At
last, performing a minimum number of operations means
less energy consumption as well as more available pro-
cessor power to perform other tasks.
The node position is the last specific criteria of the
‘‘general network’’ layer. This is a straightforward, but very
important information. Because of signal attenuation and
radio-frequency interference, nodes further away from each
other tend to require a higher amount of redundancy to
preserve a good video quality. This information becomes
even more valuable used in conjunction with the other
input parameters. For example, a much higher amount of
redundancy will be required if the network is very dense
and the nodes are far apart than if the network was not so
heavily populated.
At the end, the ‘‘general network’’ layer is responsible
for the integration of the SHIELD mechanism for the
VANETs. In this layer, all the network related issues are
tackled, allowing the proposed mechanism to be tailored
specifically for this type of network. This provides higher
performance and better QoE results.
3.2.2 The ‘‘video details’’ criteria
Besides the network conditions, the video characteristics are
also important to define a precise amount of redundancy. In
SHIELD hierarchical fuzzy system the ‘‘video details’’
0.4 0.6 0.8 1.0 1.2
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
0.4 0.6 0.8 1.0 1.2
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2
(a) (b)
Fig. 3 a Convex hull and bapproximate convex hull
0
100
200
300
400
500
600
700
800
900
1000
0 10 20 30 40 50 60 70 80 90 100
Number of operations
Number of nodes
O(n)
O(nlog(n))
O(n2)
QuickHull, worst case (30 nodes)
QuickHull, average case (30 nodes)
BFP (30 nodes)
QuickHull, above average (60 nodes)
QuickHull, average case (60 nodes)
BFP (60 nodes)
Fig. 4 Complexity of QuickHull and BFP
2568 Wireless Netw (2016) 22:2563–2577
123
criteria layer is divided into two specific components,
namely ‘‘motion activity’’ and ‘‘video characteristics’’.
The motion activity parameter is defined by the com-
bined use of temporal intensity and spatial complexity. The
temporal intensity in the SHIELD mechanism is given by
the motion vector (MV) details. The MV in the video
sequences complies with the classical mechanics concept
of vector-oriented motion model. All moving objects are
described as a simple sequence of small translations on a
plane. This was transposed to the Moving Picture Experts
Group (MPEG) standard as the movement of macroblocks
from one position, in any given frame, to another position,
in the next one. Since the MPEG standard allows the use of
distinct macroblock sizes, the SHIELD mechanism com-
putes the area of each macroblock and uses the number of
pixels that are being moved. This enables a better repre-
sentation of the intensity of the motion in arbitrary
resolutions.
In addition, the Euclidean distance of each MV is also
computed, resulting on how far each and every mac-
roblock is being moved. This information gives more
precise results than just counting the number of motion
vectors. All the input parameters are normalized, allowing
the protection of videos with arbitrary resolutions on the
fly. Following the same idea as presented before, an
exploratory analysis using hierarchical clustering is per-
formed to find the best classes that represent the temporal
intensity. After finding the classes, the fuzzy set and the
membership function can be defined. Finding the best-
fitted membership function is a complex and problem-
dependent task [36], being difficult to attain the optimal
solution. For this reason, piecewise linear functions are
desired. These functions are formed of straight-line sec-
tions and because of that, provide efficient computational
operations.
As previously mentioned, the spatial complexity is also
used to quantify the amount of the motion activity. This
parameter represents the difference of the static informa-
tion that the actual frame has when compared to the one
before. One way to compute this value is using the Sum of
Absolute Differences (SAD) [37]. This process, however,
compares each and every pixel of both frames resulting in a
very complex and time-consuming operation. Taking this
into consideration, the SHIELD mechanism uses the nor-
malized frame size to the same end. This enables a much
faster operation and, on top of that, it also allows the use of
arbitrary video resolutions.
The same process used to find the different classes in the
temporal intensity is also used to define the clusters here.
This means that, once all the frame sizes are normalized, an
exploratory analysis is performed to divide the data into the
most homogeneous groups. After that, using the linkage
distance between the clusters was possible to separate them
into five distinct groups, namely ‘‘very small’’, ‘‘small’’,
‘‘medium’’, ‘‘large’’, and ‘‘very large’’.
With the definition of the fuzzy sets completed, the
fuzzy rules must be designed. As mentioned before, this is
an intricate task, which requires a jointly knowledge of the
network details, VANETs properties, and video charac-
teristics. Aiming to reduce this complexity during the
design phase of the rules, as well as to have a better per-
formance in real-time, the SHIELD mechanism uses HFS.
This layered system allows handling fewer input parame-
ters at the same time. At the end, the result is a system with
a small number of simple rules, which lead to better
performance.
The last step of the offline process is to load all the fuzzy
sets and rules in the Fuzzy Logic Controller (FLC). Unlike
genetic algorithms or neural networks, the FLC does not
require an online training or a period of convergence,
making it an appropriate engine for real-time control. This
process has to be performed just once, during the system
bootstrap period. After that, all the functions can be
accessed in real-time. This provides the SHIELD mecha-
nism to ascertain the best-fitted QoE-aware amount of
redundancy according to each video sequence that is being
transmitted in the VANET environment.
4 Performance evaluation and results
The primary goal of the SHIELD mechanism is to enhance
the QoE, while avoiding any unnecessary network over-
head. In doing that, it improves the end-users satisfaction
and preserves the already scarce wireless resources at the
same time.
4.1 Experiment settings
In order to better characterize the performance of the
proposed mechanism two very distinct environments were
assessed: urban and highway. Each of these surroundings
features a variety of unique challenges. In the urban
environment, there are buildings and many other structures
that will affect the signal propagation. On the other hand, in
the highway environment there is much more free space,
which facilitates the signal propagation. Besides that, the
mobility patterns are also very distinctive. The urban sce-
nario presents a lot of driving options, such as avenues and
streets close to each other. On the highway is quite dif-
ferent, as there are no crossroads and just a few exits and
entrances. In addition, the speed of the vehicles has very
particular properties in each one of these environments. In
the urban case, the velocity usually is between 20 km/h and
60 km/h, and it changes frequently due to traffic lights,
speed bumps, and crosswalks. Meanwhile, on the highway,
Wireless Netw (2016) 22:2563–2577 2569
123
the speed variance is very low, staying consistently from 80
km/h to 120 km/h.
Taking account of all such differences, the Network
Simulator 3 (NS-3) [38] was used to perform the experi-
ments; both environments were simulated in a variety of
situations. Several configurations are shared, such as the
wireless and network technology, as well as the video
content and parameters. All videos were sent using Evalvid
Tool [39] and encoded with H.264, GoP length of 19:2.
Additionally, three different resolutions were used, namely
1080p, 720p, and SVGA. For each resolution, 10 videos
were chosen to be transmitted [40]. These real video
sequences cover different content of commonly viewing
material. The videos also have luminance and colour stress,
still and cut scenes, as well as distinct motion intensities
and several levels of distortions. A multi-flow scenario was
adopted. This means that up to 10 videos are transmitted
simultaneously
1
. All the receiver nodes are enabled with
Frame-Copy error concealment, meaning that each lost
frame will be replaced by the last good one.
Another feature that is the same for both environments is
the wireless standard adopted: IEEE 802.11p Wireless
Access for Vehicular Environments (WAVE) [41]. The
communication is Vehicle To Vehicle (V2V), because it
does not require a pre-existing infrastructure. Moreover,
this type of communication is envisaged as the next gen-
eration of connected cars, providing a mesh-network based
communication system, where each vehicle is able to both
send and receive information. Additionally, the routing
protocol Cross-Layer, Weighted, Position-based Routing
(CLWPR) [42] was adopted due to its position-based
characteristics. This protocol uses mobility details acquired
from the nodes to better adapt itself for a particular
VANET environment.
The mobility traces for both environments were gener-
ated using the Simulation of Urban MObility (SUMO)
[43]. This tool uses real map clippings to produce the
traces. Several details are taken into consideration, such as
routes, roundabouts, driving patterns, and traffic lights. For
the urban environment, a clipping of 2 92 km of the
Manhattan borough (New York City) was used. This
environment was simulated with up to 360 vehicles at
speeds ranging from 20 and 60 km/h. Despite the name,
SUMO can also generate highway traces considering, for
example, interchange junctions (entrance and exit ramps)
and the number of lanes. To simulate this environment a
clipping of 10 km of US Interstate Highway 78 (I-78) was
used. The number of vehicles is the same, up to 360, with
velocity between 80 and 120 km/h.
Two different propagation models were used to better
represent each environment. In the highway scenario, the
logDistance propagation model was used [44]. This is
because of the open spaces and the reduced number of
sources of interference existent in this environment. This
leads to easier communication between the nodes. On the
other hand, in the urban environment there are plenty of
sources generating interference. Because of that, on top of
the logDistance model was added the Nakagami-m prop-
agation model. This allows simulating the fast fading
characteristics commonly found in this environment [45].
Table 1summarizes the simulation parameters.
Figure 5shows the steps involved in the experiment.
First of all, the mobile traces are required from the SUMO
application (1). After that, SUMO will use real map clip-
pings from the OpenStreetMap (2) to generate the traces.
The traces enable a realistic simulation, providing more
accurate results. Following this, all the information is
loaded in the SHIELD mechanism (3). Next, the proposed
mechanism will assess the network conditions (4) and
request the video to be transmitted (5). Real video
sequences are used in the experiment (6). Afterwards, the
SHIELD mechanism optimizes and secures the video
transmission against packet loss (7). The next step is to
Table 1 Simulation parameters
Parameters Value
Display sizes 1920 91080, 1280 9720, and 800 9600
Frame rate mode Constant
Frame rate 29.970 fps
GoP 19:2
Codec H.264
Container MP4
Wireless technology IEEE 802.11p (WAVE)
Communication Vehicle to vehicle (V2V)
Routing protocol CLWPR
Mobility SUMO mobility traces
Radio range 250 m
Internet layer IPv6
Transport layer UDP
Highway environment
Propagation model logDistance
Location I-78
Map size 10,000 m
Vehicles speed 80–120 km/h (50–75 mph)
Urban environment
Propagation model logDistance ?Nakagami-m
Location Manhattan borough(New York City)
Map size 2.000 m 92.000 m
Vehicles speed 20–60 km/h (12–37 mph)
1
Samples of the transmitted videos are available in http://www.
youtube.com/channel/UCsB0SdKpCKD2GS6aXzB-FUQ/videos
2570 Wireless Netw (2016) 22:2563–2577
123
deliver the video sequences to the receiver (8). At the end,
the original (9) and the transmitted (10) videos are assessed
using objective QoE metrics (11).
Five different scenarios were assessed in both urban and
highway environments. The first one is without any type of
FEC. The results of this experiment will be used as a
baseline for the others. The second scenario is the Video-
aware Equal Error Protection FEC (VaEEP) mechanism. In
this mechanism I- and P-frames are equally protected with
a fixed amount of redundancy. The Video-aware Unequal
Error Protection FEC (VaUEP) mechanism is the third
scenario. VaUEP takes into consideration the importance
of each frame type and protects I- and P-frames with a
tailored amount of redundancy. The fourth scenario is
using our previous adaptive QoE-driven COntent-awaRe
VidEo Transmission opTimisation mEchanism (CORV-
ETTE) [28], which considers several distinct video char-
acteristics as well as the network state. The fifth, and last
scenario is the proposed SHIELD mechanism.
4.2 QoE assessments
Objective metrics are desirable to assess the video quality
level because they intend to be unbiased. In addition, they
are computed through mathematical calculations, and thus,
measurable and verifiable. The PSNR is a common
objective metric to assess data fidelity. However, it is based
on a byte-by-byte comparison disregarding completely
what the information actually represents. Additionally,
PSNR does not recognize the pixel structure in the image
nor the spatial relationship between the pixels, thus, it does
not consider the visual importance of each pixel [25,46].
To increase the results reliability, two objective QoE
metrics that mimic the human visual system [47] were
employed in order to quantify how impairments are per-
ceived, namely the Structural Similarity Metric (SSIM)
[48] and the Video Quality Metric (VQM) [49]. The MSU
Video Quality Measurement Tool [50] was used in the
experiments. In the SSIM metric, the grade system goes
from zero to one, whereas the higher the value, the better
the video quality. In the VQM metric, the closer to zero the
better quality. Another important difference between these
metrics is that VQM tends to be stricter with video
impairments. Because of that, it will give worse scores to
video sequences with fewer flaws. This will produce a
higher difference of the mechanisms results in comparison
to the baseline.
Figure 6shows QoE results of the urban scenario.
(a) and (c) depict the SSIM and VQM average, respec-
tively. (b) and (d) show the QoE improvement of each
metric in comparison to the base line. In (a), it is possible
to notice that the simulation starts with a small amount of
vehicles and the QoE results, for all mechanisms, can be
considered low. This can be credited to the fact that the
network is suffering from connectivity issues, because it is
relying on very few and scattered nodes to transmit all the
video data. Even in this scenario, the SHIELD mechanism
was able to protect the most important parts of the video
sequences, producing better results. As showed in (b), this
led to an improvement of more than 90 % on the video
quality when compared to the baseline (without FEC). The
Fig. 5 Steps involved in the experiment
Wireless Netw (2016) 22:2563–2577 2571
123
second best result was the CORVETTE mechanism with
60 % of SSIM improvement.
Figure 6a also shows that the best QoE results for all
mechanisms are obtained when the network has 160 and
200 vehicles. This number of nodes provides the best
coverage of the whole area, while it does not cause
excessive interference. Because of the improved network
conditions, the baseline also has better results, thus
reducing the SSIM improvement perceived by the other
mechanisms. This situation is clearly evidenced in Fig. 6b
for 160 and 200 vehicles. On the other hand, when the
network becomes very dense, e.g., above 280 vehicles, the
mechanisms have to face increasingly degraded network
connections. Once again, the SHIELD surpassed the other
mechanisms, providing up to 160 % higher SSIM scores in
comparison to the baseline.
As mentioned before, Fig. 6c shows the VQM average
and (d) depicts the percentage of QoE improvement of the
mechanisms in comparison to the baseline. Although this
metric differs from SSIM, almost the same pattern can be
found in (c). At the beginning of the experiment, the net-
work is sparse and the videos have low quality. VQM gives
them high scores, which in this case are not good. This is
especially true for the baseline, because it does not use any
type of FEC-based mechanism to secure the transmissions.
The best-case scenario in the VQM scores is the same as in
the SSIM results, for 160 and 200 vehicles. This confirms
the notion that the videos are transmitted with better
quality with this configuration. In the same way as in the
SSIM assessment, the VQM scores demonstrate that the
SHIELD mechanism outperforms all other mechanisms.
Additionally, Fig. 6d shows a pattern similar to the
SSIM results. The highest improvements are accomplished
when the network is sparse, between 40 and 120 vehicles,
or in a very dense network, above 240 vehicles. On aver-
age, the proposed mechanism provided scores 78 % better
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
40 80 120 160 200 240 280 320 360
SSIM
Number of vehicles
SHIELD
CORVETTE
VaEEP
VaUEP
Without FEC
0
20
40
60
80
100
120
140
160
40 80 120 160 200 240 280 320 360
Percentage of improvement(%)
Number of vehicles
SHIELD CORVETTE VaEEP VaUEP
(a) (b)
0
1
2
3
4
5
6
7
8
9
10
11
12
13
40 80 120 160 200 240 280 320 360
VQM
Number of vehicles
SHIELD
CORVETTE
VaEEP
VaUEP
Without FEC
0
100
200
300
400
500
40 80 120 160 200 240 280 320 360
Percentage of improvement (%)
Number of vehicles
SHIELD CORVETTE VaEEP VaUEP
(c) (d)
Fig. 6 QoE assessment of the urban environment. aAverage SSIM. bPercentage of SSIM improvement. cAverage VQM. dPercentage of
VQM improvement
2572 Wireless Netw (2016) 22:2563–2577
123
than the baseline. Additionally, it achieved 66 and 63 %
higher marks than VaEEP and VaUEP, respectively, and
over 40 % better scores in comparison to the CORVETTE
mechanism.
In addition to the urban scenario, a highway environ-
ment was also assessed with both SSIM and VQM metrics.
Figure 7shows the QoE assessment, whereas (a) and
(c) depict the average SSIM and VQM, respectively.
(b) and (d) show the percentage of improvement achieved
in each metric by the mechanisms against the baseline. In
(a), the first thing to be noticed is that the QoE results are
closer to one another in this environment. This happens
because the network conditions are not as harsh as in the
urban scenario. At first, there are some connectivity issues
when the network is sparse, e.g., 40 vehicles. After this
threshold, a better video quality is being provided. The best
results are evidenced for 120 and 240 vehicles. In (b), it is
possible to notice that the highest improvements are
reached when connectivity issues affect the network. For
example, when the number of deployed vehicles is 40 and
80. In addition, major improvements are also perceived
when there is a higher level of interference, such as above
280 vehicles. Here again, the SHIELD mechanism out-
performs all its competitors.
As previously mentioned, the average VQM is shown in
Fig. 7c and the percentage of VQM improvement by each
mechanism is shown in (d). In (c), the results follow the
same tendency as the SSIM scores. This means that the
VQM results are also closer to one another, especially
above 120 vehicles. This is evidenced because the highway
environment is not as rough as the urban setting. In (d), it is
clear that the highest percentage of improvement is
achieved when the nodes are sparse. This means that
connectivity issues are afflicting the network, e.g., for 40
and 80 vehicles. After this threshold, the network condi-
tions improve and the enhancements provided by the
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
40 80 120 160 200 240 280 320 360
SSIM
Number of vehicles
SHIELD
CORVETTE
VaEEP
VaUEP
Without FEC
0
10
20
30
40
50
60
70
40 80 120 160 200 240 280 320 360
Percentage of improvement(%)
Number of vehicles
SHIELD CORVETTE VaEEP VaUEP
(a)
0
1
2
3
4
5
6
7
8
9
10
11
12
13
40 80 120 160 200 240 280 320 360
VQM
Number of vehicles
SHIELD
CORVETTE
VaEEP
VaUEP
Without FEC
0
100
200
300
400
500
40 80 120 160 200 240 280 320 360
Percentage of improvement (%)
Number of vehicles
SHIELD CORVETTE VaEEP VaUEP
(c)
(b)
(d)
Fig. 7 SSIM assessment of the highway scenario. aAverage SSIM. bPercentage of SSIM improvement. cAverage VQM. dPercentage of
VQM improvement
Wireless Netw (2016) 22:2563–2577 2573
123
mechanisms decrease. Nevertheless, the SHIELD mecha-
nism is able to surpass the competitors.
4.3 Network assessment
In addition to a higher video quality, to reduce the network
overhead is also desirable. This is even more critical in
wireless networks, where the resources are scarce and
unevenly distributed. In our experiments, the network
footprint is the size of all video frames transmitted after
deducting the original frame size.
Figure 8shows the network overhead of all mechanisms
in both (a) urban and (b) highway environments. The non-
adaptive VaEEP and VaUEP schemes yield a constant
network footprint in both scenarios, because they do not
adapt the amount of redundancy according to the network
conditions. As depicted in the graph, these non-adaptive
schemes add a considerably larger amount of redundancy.
On top of that, the protection is not very efficient because,
in the VaEEP case, the protection is added equally to all
video data. As highlighted before, not all video packets
need the same degree of protection. To tackle this issue
VaUEP considers the frame type to add a specific amount
of redundancy. This results in less network overhead and,
at the same time, improves the video quality.
The VaEEP mechanism does not have a standard devi-
ation because it uses a unique and pre-defined amount of
redundancy, which is applied equally in all videos. The
VaUEP mechanism also has a pre-defined amount of
redundancy, but it is not unique. This means that each type
of video frame will have a specific amount of redundancy.
Additionally, each video frame has a different size, leading
to a variation in the amount of redundancy, and thus, a
standard deviation is displayed.
The adaptive mechanisms, CORVETTE and SHIELD,
were able to produce a lower network overhead, improving
the wireless resources usage. In both mechanisms, when
the network condition is better the footprint decreases. In
the urban environment, this is evidenced when the simu-
lation has 160 vehicles. The SHIELD mechanism produces
a network overhead of only 5 %, while CORVETTE is
producing 9 %. In the highway environment, the best
conditions are experienced between 120 and 240 vehicles.
The overhead produced by the SHIELD mechanism was
between 4 and 7 %, while CORVETTE is producing
between 8 and 15 %.
On average, the SHIELD mechanism added 20 % less
overhead in the urban environment and 28 % less in the
highway scenario, in comparison to the CORVETTE
mechanism. When compared to the VaEEP and VaUEP
mechanisms, the SHIELD mechanism produced 73 and
63 % less overhead, respectively in the urban scenario. In
the highway scenario, the network overhead downsize was
81 and 73 %, respectively. In the end, the proposed
mechanism was able to produce a tailored protection,
enabling a higher video quality and lower network
overhead.
4.4 Overall results
The overall results demonstrate that the SHIELD mecha-
nism outperforms all its competitors as showed in Table 2.
This table summarizes the average SSIM, VQM, and the
network footprint. The SHIELD mechanism enables
downsizing the network footprint in both urban and high-
way environment. This stems from the fact that a tailored
amount of redundancy, based upon video characteristics
and the network conditions, is added to each video
0
10
20
30
40
50
60
70
40 80 120 160 200 240 280 320 360
Network overhead (%)
Number of vehicles
SHIELD CORVETTE VaEEP VaUEP
0
10
20
30
40
50
60
70
40 80 120 160 200 240 280 320 360
Network overhead (%)
Number of vehicles
SHIELD CORVETTE VaEEP VaUEP
(a) (b)
Fig. 8 Network footprint. aNetwork overhead of the urban scenario. bNetwork overhead of the highway scenario
2574 Wireless Netw (2016) 22:2563–2577
123
sequence. This prevents any unnecessary redundancy.
Furthermore, the proposed mechanism also enhanced the
quality of the video delivered, thus providing higher QoE
for the end-users.
5 Conclusion and future works
Following the recent growth of video transmission over
VANETs there is the need for self-adaptive QoE-driven
mechanisms to protect the packet delivery against losses.
The SHIELD mechanism is able to safeguard the most
QoE-sensitive data, which leads to a resilient video trans-
mission. This allows improving the video quality even in
networks with high mobility nodes and error-prone ten-
dency. Through an extensive set of experiments, the pro-
posed mechanism demonstrates that it is capable of
identifying, with great accuracy, several network and video
characteristics. All these details are used to shield the most
important data, which in turn, leads to a higher video
quality and an efficient use of the wireless resources.
The experimental results show that SHIELD surpasses
the adaptive and non-adaptive competitors in both video
quality improvement and network overhead downsizing. In
terms of video quality, it achieved between 13 and 62 %
SSIM improvement in the urban environment and between
12 and 45 % of SSIM improvement in the highway envi-
ronment. In the VQM assessment, the video quality
improvement was between 67 and 358 % in the urban
environment and between 57 and 297 % in the highway
scenarios. The VQM results are higher because, as
explained before, it tends to give worst scores than SSIM as
the quality decreases.
In addition to the improved video quality, the proposed
mechanism was also able to reduce the network footprint.
The overhead downsize in the urban environment is
between 20 and 70 % and in the highway scenarios is
between 27 and 81 %. This means that it was possible to
enhance the video transmission without adding unneces-
sary redundancy, saving the scarce wireless resources. As
future work, other mobility scenarios and environments are
going to be assessed, as well as additional video and net-
work parameters.
Acknowledgments This work was funded by the Brazilian National
Counsel of Technological and Scientific Development (CNPq), and
also supported by the COST Action IC1303: AAPELE—Algorithms,
Architectures and Platforms for Enhanced Living Environments and
FCT Project, MIT-Portugal Program—SusCity: Urban data driven
models for creative and resourceful urban transitions.
References
1. comScore (2013, February). Brazilian online video audience
reaches 43 million unique viewers in december 2012. Technical
report, comScore inc. http://www.comscore.com/Insights/Press_
Releases/2013/2/Brazilian_On-line_Video_Audience_Reaches_
43_Million_Unique_Viewers_in_December_2012.
2. Adobe Digital Index. (2014). U.S. digital video benchmark
report. Technical report, Adobe (Q2 2014).
3. Zhou, L., Zhang, Y., Song, K., Jing, W., & Vasilakos, A. V.
(2011). Distributed media services in p2p-based vehicular net-
works. IEEE Transactions on Vehicular Technology,60(2),
692–703.
4. Gerla, M., Wu, C., Pau, G., & Zhu, X. (2014). Content distri-
bution in VANETs. Vehicular Communications,1(1), 3–12.
doi:10.1016/j.vehcom.2013.11.001.
5. Immich, R., Cerqueira, E., & Curado, M. (2013). Cross-layer fec-
based mechanism for packet loss resilient video transmission. In
E. Biersack, C. Callegari, M. Matijasevic (Eds.), Data traffic
monitoring and analysis. Lecture notes in computer science (vol.
7754, pp. 320–336). Springer, Berlin. doi:10.1007/978-3-642-
36784-7_13.
6. Immich, R., Cerqueira, E., & Curado, M. (2014). Ensuring qoe in
wireless networks with adaptive fec and fuzzy logic-based
mechanisms. In 2014 IEEE international conference on com-
munications (ICC) (pp. 1687–1692). doi:10.1109/ICC.2014.
6883565.
7. Soldo, F., Casetti, C., Chiasserini, C., & Chaparro, P. A. (2011).
Video streaming distribution in vanets. IEEE Transactions on
Parallel and Distributed Systems,22(7), 1085–1091. doi:10.
1109/TPDS.2010.173.
8. Shen, Z., Luo, J., Zimmermann, R., & Vasilakos, A. V. (2011).
Peer-to-peer media streaming: Insights and new developments.
Proceedings of the IEEE,99(12), 2089–2109.
9. Jiang, T., Wang, H., & Vasilakos, A. V. (2012). Qoe-driven
channel allocation schemes for multimedia transmission of pri-
ority-based secondary users over cognitive radio networks. IEEE
Journal on Selected Areas in Communications,30(7),
1215–1224.
Table 2 Average SSIM, VQM,
and network footprint SHIELD CORVETTE VaUEP VaEEP Without FEC
Urban environment
SSIM 0.895 0.787 0.701 0.684 0.551
VQM 1.459 2.441 4.034 4.323 6.688
Overhead 17.333 % 21.778 % 46.660 % 65.984 % –
Highway environment
SSIM 0.911 0.809 0.744 0.729 0.627
VQM 1.328 2.095 3.414 3.728 5.281
Overhead 12.333 % 17.112 % 46.660 % 65.984 % –
Wireless Netw (2016) 22:2563–2577 2575
123
10. Bellalta, B., Belyaev, E., Jonsson, M., & Vinel, A. (2014). Per-
formance evaluation of IEEE 802.11p-enabled vehicular video
surveillance system. IEEE Communications Letters,18(4),
708–711. doi:10.1109/LCOMM.2014.022514.140206.
11. Marwaha, S., Srinivasan, D., Tham, C.K., & Vasilakos, A.
(2004). Evolutionary fuzzy multi-objective routing for wireless
mobile ad hoc networks. In Congress on evolutionary computa-
tion, 2004. CEC2004 (vol. 2, pp. 1964–1971). IEEE.
12. Zeng, Y., Xiang, K., Li, D., & Vasilakos, A. V. (2013). Direc-
tional routing and scheduling for green vehicular delay tolerant
networks. Wireless Networks,19(2), 161–173.
13. Pham, T. A. Q., Piamrat, K., & Viho, C. (2014). Qoe-aware
routing for video streaming over vanets. In 2014 IEEE 80th
vehicular technology conference (VTC Fall) (pp. 1–5). doi:10.
1109/VTCFall.2014.6966141.
14. Wu, H., & Ma, H. (2014). Opportunistic routing for live video
streaming in vehicular ad hoc networks. In 2014 IEEE 15th
international symposium on a world of wireless, mobile and
multimedia networks (WoWMoM) (pp. 1–3). doi:10.1109/WoW
MoM.2014.6919002.
15. Zhang, X. M., Zhang, Y., Yan, F., & Vasilakos, A. V. (2015).
Interference-based topology control algorithm for delay-con-
strained mobile ad hoc networks. IEEE Transactions on Mobile
Computing,14(4), 742–754.
16. Nafaa, A., Taleb, T., & Murphy, L. (2008). Forward error cor-
rection strategies for media streaming over wireless networks.
IEEE Communications Magazine,46(1), 72–79. doi:10.1109/
MCOM.2008.4427233.
17. Immich, R., Borges, P., Cerqueira, E., & Curado, M. (2015).
QoE-driven video delivery improvement using packet loss pre-
diction. International Journal of Parallel, Emergent and Dis-
tributed Systems.
18. Zhou, L., Chao, H.-C., & Vasilakos, A. V. (2011). Joint forensics-
scheduling strategy for delay-sensitive multimedia applications
over heterogeneous networks. IEEE Journal on Selected Areas in
Communications,29(7), 1358–1367.
19. Greengrass, J., Evans, J., & Begen, A. C. (2009). Not all packets
are equal, part I: Streaming video coding and sla requirements.
IEEE Internet Computing,13, 70–75. doi:10.1109/MIC.2009.14.
20. Wan, Z., Xiong, N., Ghani, N., Vasilakos, A., & Zhou, L. (2014).
Adaptive unequal protection for wireless video transmission over
IEEE 802.11e networks. Multimedia Tools and Applications,
72(1), 541–571. doi:10.1007/s11042-013-1378-z.
21. Raju, G. V. S., Zhou, J., & Kisner, R. A. (1991). Hierarchical fuzzy
control. International Journal of Control,54(5), 1201–1216.
doi:10.1080/00207179108934205.
22. Rubino, G. (2005). Quantifying the quality of audio and video
transmissions over the internet: The PSQA approach. Design and
operations of communication networks: A review of wired and
wireless modeling and management challenges. London: Imperial
College Press.
23. Asefi, M., Mark, J. W., & Shen, X. (2012). A mobility-aware and
quality-driven retransmission limit adaptation scheme for video
streaming over vanets. IEEE Transactions on Wireless Communica-
tions,11(5), 1817–1827. doi:10.1109/TWC.2012.030812.111064.
24. Naeimipoor, F., & Boukerche, A. (2014). A hybrid video dis-
semination protocol for vanets. In 2014 IEEE international
conference on communications (ICC) (pp. 112–117). doi:10.
1109/ICC.2014.6883304.
25. Huynh-Thu, Q., & Ghanbari, M. (2008). Scope of validity of
PSNR in image/video quality assessment. Electronics Letters,44,
800–8011.
26. Wang, Z., & Hassan, M. (2012). Blind xor: Low-overhead loss
recovery for vehicular safety communications. IEEE Transac-
tions on Vehicular Technology,61(1), 35–45. doi:10.1109/TVT.
2011.2172010.
27. Rezende, C., Almulla, M., & Boukerche, A. (2013). The use of
erasure coding for video streaming unicast over vehicular ad hoc
networks. In 2013 IEEE 38th conference on local computer
networks (LCN), pp. 715–718. doi:10.1109/LCN.2013.6761318.
28. Immich, R., Cerqueira, E., & Curado, M. (2015). Adaptive qoe-
driven video transmission over vehicular ad-hoc networks. In
2015 IEEE conference on computer communications workshops
(INFOCOM WKSHPS).
29. Vlavianos, A., Law, L. K., Broustis, I., Krishnamurthy, S. V., &
Faloutsos, M. (2008). Assessing link quality in IEEE 802.11
wireless networks: Which is the right metric? In IEEE 19th
international symposium on personal, indoor and mobile radio
communications, 2008. PIMRC 2008 (pp. 1–6). doi:10.1109/
PIMRC.2008.4699837.
30. Wan, Z., Xiong, N., & Yang, L. (2015). Cross-layer video
transmission over IEEE 802.11e multihop networks. Multimedia
Tools and Applications,74(1), 5–23. doi:10.1007/s11042-013-
1447-3.
31. Martini, M. G., Mazzotti, M., Lamy-Bergot, C., Huusko, J., &
Amon, P. (2007). Content adaptive network aware joint opti-
mization of wireless video transmission. IEEE Communications
Magazine,45(1), 84–90. doi:10.1109/MCOM.2007.284542.
32. Immich, R., Cerqueira, E., & Curado, M. (2014). Towards the
enhancement of uav video transmission with motion intensity
awareness. In Wireless Days (WD), 2014 IFIP.
33. Immich, R., Cerqueira, E., & Curado, M. (2013). Adaptive video-
aware fec-based mechanism with unequal error protection
scheme. In Proceedings of the 28th annual ACM symposium on
applied computing (pp. 981–988). ACM.
34. Barber, C. B.,Dobkin, D. P., & Huhdanpaa, H. (1996).The quickhull
algorithm for convex hulls. ACM Transactions on Mathematical
Software,22(4), 469–483. doi:10.1145/235815.235821.
35. Bentley, J. L., Preparata, F. P., & Faust, M. G. (1982). Approx-
imation algorithms for convex hulls. Communications of the
ACM,25(1), 64–68. doi:10.1145/358315.358392.
36. Wong, K.-W., Tikk, D., Gedeon, T. D., & Koczy, L. T. (2005).
Fuzzy rule interpolation for multidimensional input spaces with
applications: A case study. IEEE Transactions on Fuzzy Systems,
13(6), 809–819. doi:10.1109/TFUZZ.2005.859316.
37. Vanne, J., Aho, E., Hamalainen, T. D., & Kuusilinna, K. (2006).
A high-performance sum of absolute difference implementation
for motion estimation. IEEE Transactions on Circuits and Sys-
tems for Video Technology,16(7), 876–883. doi:10.1109/TCSVT.
2006.877150.
38. Henderson, T. R., Roy, S., Floyd, S., & Riley, G. F. (2006). Ns-3
project goals. In Proceeding from the 2006 workshop on Ns-2:
The IP network simulator. WNS2 ’06. ACM, New York, NY.
doi:10.1145/1190455.1190468.
39. Klaue, J., Rathke, B., & Wolisz, A. (2003). Evalvid—A frame-
work for video transmission and quality evaluation. 13th Inter-
national conference on modeling techniques and tools for
computer performance evaluation (pp. 255–272).
40. Xiph.org Video Test Media [derf’s collection]. http://media.xiph.
org/video/derf/.
41. Jiang, D., & Delgrossi, L. (2008). IEEE 802.11p: Towards an
international standard for wireless access in vehicular environ-
ments. In IEEE vehicular technology conference, 2008. VTC
Spring 2008. (pp. 2036–2040). doi:10.1109/VETECS.2008.458.
2576 Wireless Netw (2016) 22:2563–2577
123
42. Katsaros, K., Dianati, M., Tafazolli, R., & Kernchen, R. (2011).
Clwpr—A novel cross-layer optimized position based routing
protocol for vanets. In 2011 IEEE vehicular networking confer-
ence (VNC) (pp. 139–146). doi:10.1109/VNC.2011.6117135.
43. Behrisch, M., Bieker, L., Erdmann, J., & Krajzewicz, D. (2011).
Sumo–simulation of urban mobility. In The third international
conference on advances in system simulation (SIMUL 2011),
Barcelona.
44. Mittag, J., Papanastasiou, S., Hartenstein, H., & Strom, E. G.
(2011). Enabling accurate cross-layer PHY/MAC/NET simula-
tion studies of vehicular communication networks. Proceedings
of the IEEE,99(7), 1311–1326. doi:10.1109/JPROC.2010.
2103291.
45. Taliwal, V., Jiang, D., Mangold, H., Chen, C., & Sengupta, R.
(2004). Empirical determination of channel characteristics for
DSRC vehicle-to-vehicle communication. In Proceedings of the
1st ACM international workshop on vehicular ad hoc networks.
VANET ’04 (pp. 88–88). ACM, New York, NY. doi:10.1145/
1023875.1023890.
46. Winkler, S., & Mohandas, P. (2008). The evolution of video
quality measurement: From PSNR to hybrid metrics. IEEE
Transactions on Broadcasting,54(3), 660–668. doi:10.1109/
TBC.2008.2000733.
47. Chikkerur, S., Sundaram, V., Reisslein, M., & Karam, L. J.
(2011). Objective video quality assessment methods: A classifi-
cation, review, and performance comparison. IEEE Transactions
on Broadcasting,57(2), 165–182. doi:10.1109/TBC.2011.
2104671.
48. Wang, Z., Bovik, A. C., Sheikh, H. R., & Simoncelli, E. P.
(2004). Image quality assessment: From error visibility to struc-
tural similarity. IEEE Transactions on Image Processing,13(4),
600–612. doi:10.1109/TIP.2003.819861.
49. Pinson, M. H., & Wolf, S. (2004). A new standardized method for
objectively measuring video quality. IEEE Transactions on
Broadcasting,50(3), 312–322.
50. Vatolin, D., Moskin, A., Pretov, O., & Trunichkin, N. Msu video
quality measurement tool.http://compression.ru/video/quality_
measure/video_measurement_tool_en.html.
Roger Immich is a Ph.D. stu-
dent at Department of Informat-
ics Engineering, University of
Coimbra, Portugal. He received
his M.Sc. degree in Computer
Science from Federal University
of Santa Catarina, Brazil in 2006.
He worked for several years as
assistant professor at Faculty of
Technology SENAI, and also as
a Team Leader in the private
sector. His research involves
Multimedia, Quality of Experi-
ence, Vehicular Ad-hoc Net-
works, and Wireless Networks.
Eduardo Cerqueira received
his Ph.D. in Informatics Engi-
neering from the University of
Coimbra, Portugal (2008). He is
an associate professor at the
Faculty of Computer Engineer-
ing of the UFPA in Brazil. His
publications include 5 edited
books, 5 book chapters, 4
patents and over than 150
papers in national/international
refereed journals/conferences.
He is involved in the organiza-
tion of several international
conferences and workshops,
including Future Multimedia Networking (IEEE FMN), Future
Human-centric Multimedia Networking (ACM FhMN), ICST Con-
ference on Communications Infrastructure, Systems and Applications
in Europe (EuropeComm), Latin America Conference on Communi-
cations (IEEE LATINCOM) and Latin American Conference on
Networking (IFIP/ACM LANC). He has been serving as a Guest
Editor for 5 special issues of various peer-reviewed scholarly jour-
nals. His research involves Multimedia, Future Internet, Quality of
Experience, Mobility and Ubiquitous Computing.
Marilia Curado is a Tenured
Assistant Professor at the
Department of Informatics
Engineering of the University of
Coimbra, Portugal, from where
she got a Ph.D. in Informatics
Engineering on the subject of
Quality of Service Routing, in
2005. Her research interests are
Quality of Service, Quality of
Experience, Energy efficiency,
Wireless Networks, Mobility,
Cloud Systems, and Software
Defined Networks. She is the
coordinator of the Laboratory of
Communications and Telematics of the Centre for Informatics and
Systems of the University of Coimbra. She has been general and TPC
chair of several conferences and belongs to the editorial board of
Elsevier Computer Networks. She has participated in several national
projects, in Networks of Excellence from IST FP5 and FP6, in the IST
FP6 Integrated Projects, EuQoS and WEIRD, and on ICT FP7
STREPs MICIE, GINSENG and COCKPIT. She acts regularly as an
evaluator for EU projects and proposals.
Wireless Netw (2016) 22:2563–2577 2577
123
A preview of this full-text is provided by Springer Nature.
Content available from Wireless Networks
This content is subject to copyright. Terms and conditions apply.
