Crowdsourcing Location Sensitive Data for Dynamic Scenario by
Adaptive Role Assignment
Anubhuti Garg and Amiya Nayak
School of Electrical Engineering and Computer Science, University of Ottawa, Ontario, Canada
Keywords:
Participatory Sensing, Localization, Collaborative.
Abstract:
The existing technique for performing crowdsourced, location-based sensing activity minimizes energy con-
sumption by eliminating the use of GPS by some devices. For this, server detects a set of participants for the
role of broadcaster which must turn-on their GPS to collect location information and broadcast it to neighbour-
ing devices for their position calculation. However, if new devices join the region then they cannot participate
in the ongoing sensing task until next localization phase when server reassigns role to all participants. In
addition to this, if devices leave the region then their neighbouring devices may require a change of role. The
current work does not provide solution to such dynamic scenarios. We provide time and energy efficient ap-
proach to allocate role adaptively to participants when they join or leave the region of interest. For this, we
propose incremental algorithms to assign role for the new participants joining the region and for modifying the
roles of existing participants when some devices leave the region. This also eliminates the need for rerunning
the role-assignment algorithm over the entire set of participants for every insertion and deletion. The proposed
solutions are capable of saving 95-99.9% of the role assignment time without compensating energy needs.
1 INTRODUCTION
Mobile phones have become an indispensable part
of our lives and this has attracted researchers to har-
ness its data sensing capabilities and extract valuable
knowledge. Most smartphones are embedded with
rich set of sensors such as accelerometer, GPS, gy-
roscope, microphone, camera and interfaces such as
WiFi, bluetooth and other technologies (Lane et al.,
2010). This has lead to number of exciting applica-
tions based on mobile phone sensing.
In this paper, our focus is on crowdsourcing data
for participatory sensing system. In such a system,
participants actively participate in sensing activity
and collaborate to accomplish a given task (Macias
et al., 2013). Participatory sensing supports various
applications ranging from heath services to environ-
ment monitoring, most of which are dependent on
location information. For accurate location informa-
tion, devices depend on GPS which is a major source
of power depletion in cell phones. Therefore, re-
searchers focus on providing alternatives to GPS us-
age. In (Song et al., 2014), authors provide a device
to device localization scheme to relieve some devices
from using GPS and thereby saving their phone’s en-
ergy. It takes into account mobility and relative posi-
tioning of devices. For accuracy, some devices need
to turn on GPS and others depend on them for calcu-
lation of their location.
Figure 1: System Architure.
Garg, A. and Nayak, A.
Crowdsourcing Location Sensitive Data for Dynamic Scenario by Adaptive Role Assignment.
DOI: 10.5220/0006436000450054
In Proceedings of the 14th International Joint Conference on e-Business and Telecommunications (ICETE 2017) - Volume 6: WINSYS, pages 45-54
ISBN: 978-989-758-261-5
Copyright © 2017 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
45

Our research is motivated by the application
framework shown in Fig. 1 for collaborative sens-
ing. Similar framework was also used by authors in
(Garg et al., 2017) and (Wang et al, 2016). The task
publisher sends the sensing task to server which for-
wards the same to participating smartphones. The
tasks are such that sensing requires location of the de-
vices. These participants are assigned a role for some
time period by the server after which roles are reas-
signed according to the updated location information.
There are three categories of role that can be assigned
-
• Broadcaster: Its function is to obtain location us-
ing GPS then broadcast this location and move-
ment information to the surrounding participants;
• Location Information Receivers (LIR): They rely
on broadcasters to calculate their location using
device-to-device localization method;
• Normal Participants (NP): They do not receive
any broadcast from their surrounding so, they de-
pend on GPS to obtain location information.
These roles last for certain period of time then server
assigns new roles according to their updated location.
The authors in (Garg et al., 2017) provide an ef-
ficient energy consumption model for participants in
three roles and a sorting based method used by the
server to assign role for each of the participating de-
vices. However, there is an inherent assumption of
fixed number of participants. In reality, the region
can have frequent or infrequent updates because of
the dynamic nature of smartphone users. For instance,
consider an application that maintains gas prices at
different locations in a locality. For this, participat-
ing smartphones upload prices for the gas and loca-
tion whenever they make use of it. The gas stations
on highways are generally used by tourists more. In
such scenarios, the application must ensure that server
is capable of assigning role to new participants and
adapt to changes quickly and efficiently.
Updates are usually collected and applied period-
ically due to which these devices cannot participate
in the ongoing sensing task and have to wait for next
assignment round. However, we make it possible to
assign role to such new devices before the task arrives
without re-running greedy (Wang et al., 2016) or sort-
ing based (Garg et al., 2017) algorithm. Due to large
number of participating devices, it is desirable to as-
sign role incrementally. This is the first study so far
where we consider devices joining or leaving the re-
gion on fly. If a device joins the existing set of partic-
ipants then the role is assigned based on its location,
and in case device leaves the region then we reassign
the role of its neighbouring devices only.
The rest of the paper is organized as follows. Sec-
tion II presents related work. In Section III, we dis-
cuss various scenarios that might be possible when
a device joins or leaves the region. We also present
incremental insertion and incremental deletion algo-
rithms to consider all cases. In Section IV, we present
a model for the adaptive approach, taking into account
both insertion and deletion algorithms. The exper-
imental results are presented in Section V followed
by conclusion and future work in Section VI. In the
paper, we have used participants, devices, nodes and
smartphones interchangeably.
2 RELATED WORK
Participatory sensing has shown its great potential in
numerous application domains such as health care,
environment monitoring, transportation, social net-
works, safety, industrial monitoring and maintenance,
academia and government agencies. For example,
AndWellness (Hicks et al., 2010) was developed as
a personal behavioral and contextual data collec-
tion system to record and monitor participants’ daily
habits. The system could be even deployed to assess
HIV+ patients through behavior and emotional dis-
tress. Similarly, Biketastic (Reddy et al., 2010) is an-
other system that collects route experience of bikers
such as terrain, noise level, scenery image. The data
was collected periodically after every one second us-
ing GPS. It used accelerometer to sense noise level
and roughness. The information of different routes
was made available to cyclists in order to facilitate
them for choosing right path.
Participatory sensing involves participation of all
devices. First, the sensing task is initialized to deter-
mine the goal, what and how to sense. Then, the task
is passed on to the users to collect data which can be
location-based or logged or manual information. The
data is then transferred to the server or cloud for pro-
cessing (Goldman et al., 2009).
Energy is one of the most important issues that
have been considered in the study of participatory
sensing. In (Song et al., 2014), authors provide an
energy-efficient participant selection algorithm based
on constrained optimization problem and quality of
information(QoI) which includes sensing region, time
period, data granularity and quantity. They propose a
behavioral model to find relationship between resid-
ual energy and willingness to participate so as to know
before hand which participant will deny participating
in the sensing activity. However, they consider only
a subset of participants for performing required sens-
ing task. PSense (Baier et al., 2012) is another ap-
WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems
46

plication proposed to reduce energy consumption of
mobile devices. They make use of the adaptive po-
sitioning mechanism and short range communication
to exchange position related information. The server
sends the set of locations that are needed to be sensed.
Each mobile device then periodically fix its location
to cover all queried regions. Jigsaw (Lu et al., 2010)
provided an energy efficient sensing engine developed
to continuously monitor human activities and con-
texts. But, the authors do not consider sensing tasks
based on location of the devices.
Most sensing techniques which need position in-
formation rely on GPS. However, it drains suffi-
cient power of the phones. Hence, many alterna-
tives to GPS have been proposed compromising ac-
curacy of the devices. In (Shafer et al., 2010), authors
present an indoor WLAN-localization method using
accelerometer. This is an energy-efficient technique
but can work only in indoor environment. Some lo-
calization techniques are based on Bluetooth technol-
ogy (Johnson et al., 2012). In (Kumar et al., 2013), a
method for localization using location beacon is pro-
vided which requires fixed or mobile beacons to esti-
mate position. (Song et al., 2014) proposed a device-
to-device localization method which uses propagation
model of wireless signals. The movements of de-
vices are calculated by inertial sensors using step-up
method and change in distance between devices are
modelled by change in signal strength. This method
was deployed in (Wang et al., 2016) for collaborative
outdoor localization. A server selects set of devices
which must turn on GPS while neighbouring devices
rely on them for calculating location using device-to-
device localization.
Authors in (Wang et al., 2016) provided a greedy
algorithm which we refer to as GBS (Greedy based
Broadcaster Set selection) for the role assignment.
The basic idea was to select optimal set of broadcast-
ers. The algorithm checks every participant in each
iteration and assigns the role of broadcaster to the one
which minimizes system energy the most. It keeps
iterating until no participant can be chosen for broad-
caster’s role that can minimize system energy further.
Devices which are not close to any other device are
selected as normal participant, and rest of them are
chosen for the role of location information receivers.
In (Garg et al., 2017), authors proposed a sorting
based algorithm which they refer to as SBS (Sorting
based Broadcaster Set selection) for the selection of
broadcasters. In this, they introduced the following
terms:
• Connectivity(κ) to represent number of devices
within WiFi range of participant, and
• Local Connectivity(κ
0
) to denote number of de-
vices that participant would contribute to LIR set.
The algorithm first sorts participants on the basis of
their connectivity. For every iteration, it sets an up-
per bound to minimize the search space for selecting
the next broadcaster. Within this range, the node with
highest local connectivity and that minimizes energy
of system is chosen as next broadcaster. Location in-
formation receivers and normal participants are cho-
sen in the same way as in the greedy approach.
The proposed solution for adaptive changes can
be applied to both the algorithms. We have compared
our algorithm with sorting based approach as its ef-
fectiveness over greedy has been proven in (Garg et
al., 2017). However, results would remain same if we
compare with greedy solution as well.
3 INCREMENTAL OPERATIONS
In this approach, the following two operations are
considered:
• Insertion - new participant joins the region of in-
terest,
• Deletion - existing participant leaves region of in-
terest.
The role of new mobile device being inserted is
dependent on its location. For instance, if the device
is not within WiFi range of any other device then it
act as a normal participant. Similarly, the change of
role for the existing devices due to exit of participants
is affected only within the WiFi reception range of the
device being deleted.
In the following sections, we discuss various cases
due to insertion and deletion of a participant and sub-
sequently provide an algorithm for each case.
3.1 Insertion
When a participant p joins the region, new connec-
tions may be established, but none is removed. Fol-
lowing cases can occur:
• (Noise)
If the new participant cannot become part of any
broadcaster, i.e., it is not close enough to any
broadcaster to receive its WiFi signal, then it be-
comes a normal participant. Fig. 2(case 1) depicts
similar case. A node x labelled as p is being in-
serted to an existing set of devices. However, its
WiFi range does not cover any node, hence as-
signed a role of a normal participant.
• (Creation)
If the new participant’s reception range covers few
Crowdsourcing Location Sensitive Data for Dynamic Scenario by Adaptive Role Assignment
47

Table 1: List of notations.
Notation Explanation
M Set of smartphones
B
t
1
t
2
Set of broadcasters during [t
1
,t
2
]
b
m
Boolean to indicate if smartphone m is selected as broadcaster
br Boolean to indicate if smartphone m is selected as LIR
BLIR
t
1
t
2
Set of LIRs for each broadcaster for the interval [t
1
,t
2
]
κ Connectivity of participant
κ
0
Local Connectivity of a participant
P
t
1
t
2
Physical connectivity matrix for interval [t
1
,t
2
]
P
local
Local physical connectivity matrix
I Set of new participants being inserted
D Set of participants to be deleted
e
b
Energy of broadcaster
e
n
Energy of normal participant
e
l
Energy of LIR
E
t
1
t
2
Energy consumed during [t
1
,t
2
]
δ Threshold for the change in database(in %)
normal nodes then it is eligible to become a broad-
caster. It can then switch on its GPS, collect sens-
ing information and update server through cellular
network before task deadline. Fig. 2(case 2) illus-
trates this case. The green dots are used to depict
normal nodes. When a new node p covers two
normal nodes within its WiFi range, it is assigned
a role of a broadcaster, and the covered nodes are
reassigned the role of LIR.
• (Absorption)
If the new participant is close enough to any of
the existing broadcasters then it becomes part of
it and act as a LIR node. Fig. 2(case 3) shows
absorption of newly inserted node, p. It is close
enough to send and receive signals from an exist-
ing broadcaster depicted by red colour. Hence, it
gets absorbed and is assigned a role of LIR.
3.1.1 Incremental Insertion Algorithm
In this subsection we provide algorithm for incremen-
tal insertion to consider the cases discussed above.
The for loop from Step 1 to 19 iterates through entire
set of new participants joining the region. Each of the
parameter, E
1
, E
2
, and E
3
denotes energy consumed
if participant is selected as broadcaster (Step 3), LIR
(Step 4) or NP (Step 5) respectively. E
1
is computed
using the local connectivity of new participant that is,
number of normal participants that it can cover (Step
3). E
2
is updated when the participant is within WiFi
range of any broadcaster. The flag is then set to 1 and
energy of LIR (e
l
) is added to E
2
(Step 6 to 9). E
3
is
obtained by simply adding energy of NP (e
n
) in case
node is selected as NP. The minimum among the three
energies is used to assign role to the new node (Step
11 to 17). In case new node is assigned the role of
broadcaster then, the normal participants that it cov-
ers change the role to LIR and boolean vector(br) is
updated. If br(i) is set to unity then it represents that
i
th
participant is assigned a role of LIR and if it is set
to zero then participant can be broadcaster or normal
participant.
3.2 Deletion
As opposed to insertion, when a participant p leaves
the group, connections or role of existing devices
might change. The trickiest case is when a broad-
caster leaves the region. We discuss all possible cases
below when deleting a node x, labelled as p:
• (Removal)
If LIR exits such that its broadcaster still covers
a set of nodes then it simply updates its broad-
caster with sensing data collected so far and asks
to remove it from its database. LIR is removed
without affecting roles of any other device. Fig.
3(case 1.1) shows a similar scenario. Deleting
node p does not affect role of any other node. The
broadcaster continues to cover large set of devices
(LIR).
Similarly, if normal node decides to go out of
the region of interest, then it gets deleted with-
out affecting any other participant. When server
does not listen any update then it is considered
to be deleted, and the database is updated. Fig.
3(case 1.2), shows a similar scenario. The node
p is initially assigned a role of normal participant
(coloured green to depict its role). However, its
WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems
48

Figure 2: Different cases of the Insertion algorithm.
exit from the region does not affect role of any
other device.
• (Reduction)
Once again, we consider the case when LIR leaves
the region. We discussed previously how exit of
LIR does not affect role of any other participant.
The main idea for the role assignment is to mini-
mize energy needs. If reassigning roles minimizes
energy further, then it is a preferred choice. For
instance, in Fig. 3(case 2), when node p gets
deleted, it leaves broadcaster with no LIR node.
Algorithm 1: Incremental Insertion Algorithm.
Input: Initial set of participants: M,
Set of broadcaster: B
t
1
t
2
,
Energy E
t
1
t
2
for M,
Boolean vector: br,
Set of new participants being inserted, I
Output: Role for each node in I
1: for each x ∈ I do
2: Init f lag;
3: E
1
= E
t
1
t
2
+ e
b
+ κ
0
x
× (e
l
− e
n
);
4: E
2
= E
t
1
t
2
;
5: E
3
= E
t
1
t
2
+ e
n
;
6: if x is physically connected to any b ∈ B
t
1
t
2
then
7: f lag = 1;
8: E
2
= E
2
+ e
l
;
9: end if
10: if f lag == 1 && E
2
≤ E
1
then
11: Assign x the role of LIR;
12: else if (( f lag == 1 && E
1
≤ E
2
&& E
1
≤
E
3
) k ( f lag == 0 && E
1
≤ E
3
)) then
13: Assign x the role of Broadcaster;
14: Update br;
15: else
16: Assign x role of NP;
17: end if
18: Update E
t
1
t
2
;
19: end for
In such case, it is better to reassign role to this
broadcaster as normal participant because e
b
> e
n
.
• (Deletion)
In this, we consider the case when broadcaster
goes out of the region of interest. Since it knows
the location of every neighbouring participant (its
LIR devices), it runs sorting based or greedy al-
gorithm over its LIRs and reassign roles to them.
New broadcasters are chosen if they can cover
LIRs; otherwise, node is reassigned role of nor-
mal participant. Fig. 3(case 3.1), depicts the case
when a node can be chosen as broadcaster which
covers all LIRs of p. However in Fig 3(case 3.2),
when a broadcaster node p decides to leave the
region, it leads to formation of two other broad-
casters to cover its LIR nodes.
The broadcaster being deleted shares its data, task
and deadline with newly assigned broadcasters
and normal participants. The information of the
node being deleted would be updated to the server
at the end of the task.
3.2.1 Incremental Deletion
In this subsection, we provide an algorithm for incre-
mental deletion (Algorithm 2) to consider all the cases
discussed above. For each deletion (Step 1), we find
their existing role. If this participant x, happens to be
a NP then it is simply removed(Step 39). If x was
assigned a role of LIR then we remove it from LIR
set by replacing one with zero in br vector(Step 3).
We find its corresponding broadcaster using BLIR
t
1
t
2
matrix(Step 4). The BLIR
t
1
t
2
maintains the LIRs that
broadcaster covers. Whenever a device is chosen as
broadcaster, its index is appended in matrix B
t
1
t
2
. Its
corresponding set of LIRs represented by Boolean
vector, is appended in Boolean matrix, BLIR
t
1
t
2
. The
bit at (i, j) is set to unity when j
th
device is chosen as
LIR for i
th
broadcaster.
In case, the broadcaster of x no longer covers
any LIR then its role is changed to normal partici-
pant(Step 6 to 9). If participant, x, was assigned the
Crowdsourcing Location Sensitive Data for Dynamic Scenario by Adaptive Role Assignment
49

Algorithm 2: Incremental Deletion Algorithm.
Input: Initial set of participants: M,
Set of broadcaster: B
t
1
t
2
,
Boolean vector: br,
Set of LIRs for each broadcaster: BLIR
t
1
t
2
Physical connectivity matrix: P
t
1
t
2
Energy E
t
1
t
2
for M,
Set of participants to be deleted, D
Output: Role for each node in M
1: for each x ∈ D do
2: if br(x)==1 then
3: br(x) = 0;
4: b = x
0
s broadcaster;
5: BLIR
t
1
t
2
(b, x) = 0;
6: if sum(BLIR
t
1
t
2
(b)) == 0 then
7: b becomes NP;
8: Remove b from B
t
1
t
2
;
9: end if
10: else if x exists in B
t
1
t
2
then
11: i=index of x in B
t
1
t
2
;
12: Initialize P
local
;
13: for each j : BLIR
t
1
t
2
(i, j) = 1 do
14: temp=BLIR
t
1
t
2
(i) ∧ P
t
1
t
2
(x);
15: temp( j)=0;
16: Insert temp in P
local
;
17: end for
18: κ
0
x
= sum(BLIR
t
1
t
2
(i));
19: Sort P
local
in descending order of κ;
20: iter = number of rows in P
local
;
21: k = 1;
22: while ρ = 0 && k = iter do
23: m =Index of P
local
(k);
24: κ
0
= sum(BLIRt
1
t
2
(i) ∧ P
local
(k);
25: if κ
0
== 0 then
26: br(m) = 0;
27: Assign role of NP to m;
28: κ
0
x
= κ
0
x
− 1;
29: else
30: br(m) = 0;
31: B
t
1
t
2
∪ {m};
32: BLIR
t
1
t
2
= BLIR
t
1
t
2
∪ P
local
(k);
33: BLIR
t
1
t
2
(i) = BLIR
t
1
t
2
(i) ⊗
P
local
(k);
34: BLIR
t
1
t
2
(i, m) = 0;
35: κ
0
x
= κ
0
x
− κ
0
− 1;
36: end if
37: end while
38: else
39: Remove x from NP set;
40: end if
41: end for
role of broadcaster then we obtain a local physical
connectivity matrix, P
local
(Step 12). Our aim is to
find new set of broadcasters that can cover its LIRs.
For this, every entry of P
local
is obtained by the oper-
ation of Boolean AND over LIR set of x and physical
connectivity of its LIRs(Step 13 to 17). Next, we ob-
tain local connectivity of x , represented by κ
0
x
(Step
18). We sort P
local
in the descending order of partic-
ipant’s connectivity(Step 19). We then select partici-
pants from P
local
that can cover LIRs of x(Step 30 to
35). The local connectivity of each participant is ob-
tained by the Boolean AND operation(Step 24). In
case they do not cover any LIR then they are removed
from LIR set (Step 26) and assigned role of normal
participant (Step 27). The local connectivity of x is
decremented by 1 whenever its LIR is assigned a role
of normal participant(Step 28). If participant covers
some LIRs of x then it is added to the broadcaster
set(Step 31), and corresponding LIR set is added to
BLIR
t
1
t
2
(Step 32). With this, we update the local con-
nectivity of x by removing participants that have been
covered using boolean XOR operation(Step 33) and
removing the participant from x
0
s LIR list that is re-
cently chosen for broadcaster role(Step 34) and global
LIR list, br. The κ
0
x
is then updated(Step 35).
4 PROPOSED MODEL
In dynamic scenario, we consider participants join-
ing or leaving the region on fly and aim to assign role
without rerunning greedy or sorting based algorithms
for each insertion or deletion. However, our aim is
to minimize energy. The proposed incremental algo-
rithms provided in Algorithm 3, 2 do not provide op-
timal set of broadcasters. Hence, there is a need to re-
run the algorithm when energy consumption becomes
too high.
We set a threshold, δ to check the change in
database. If change in database is less than δ then
it is better to use the incremental technique to assign
role (Step 4); otherwise, we recommend to rerun SBS
or GBS algorithm (Step 6).
The percentage change in database can be calcu-
lated by following:
% Change in DB =
NewData − OldData
OldData
× 100.
The proposed model is also depicted by Fig. 4.
The actual SBS (or GBS) algorithm is applied to
the original database to assign role to each partici-
pant(Step 1). Then, we use incremental insertion and
deletion algorithm to adapt new changes to the dataset
if the change is less than some threshold, δ(Step 3,4)
otherwise rerun SBS/GBS(Step 6).
WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems
50

Figure 3: Different cases of the Deletion algorithm.
Algorithm 3: Adaptive Algorithm.
1: Apply SBS/GBS on the original set of partici-
pants.
2: for each x inserted or deleted do
3: if % change in database < δ then
4: Apply incremental insertion or deletion
algorithm to assign role to x
5: else
6: Rerun SBS/GBS algorithm
7: end if
8: end for
Figure 4: Proposed Model for Adaptive Algorithm.
5 PERFORMANCE EVALUATION
We evaluate the performance of proposed adaptive
strategy using synthetic dataset of size 100, 300, 500
and 600. For each participant, we generate random
positions in terms of x and y coordinate confined in
area of 500 × 500m
2
. For each case, a distance ma-
trix is generated to calculate distance between every
device to derive physical connectivity matrix. All pa-
rameters are set with same values as used in (Wang et
al., 2016). We have used Matlab for all simulations
and experiments. We have not experimented with
GBS algorithm as its performance is already evalu-
ated in (Garg et al., 2017), and the results remain
same if it is compared with incremental GBS. Also,
we have used energy model presented in (Garg et al.,
2017) for experiments based on energy consumption.
In first subsection, we discuss results of the insertion
algorithm which is followed by the results of the dele-
tion algorithm.
5.1 Incremental Insertion
In each of the results, we call the proposed algorithm
as Incremental Insertion and SBS for rerunning SBS
algorithm for each insertion. Each of the following
results was obtained as an average of 25 runs on 25
datasets, each of the size {100, 300, 500, 600}.
In the first experiment, we aim to evaluate the time
taken for assigning roles by the proposed incremental
insertion and SBS algorithm. For this, the number of
new participants inserted is equivalent to 5% of the
dataset size. The sorting based algorithm has to be
rerun for every insertion. Fig. 5 clearly depicts that
our algorithm outperforms SBS.
In the next experiment, we evaluate the impact on
energy consumption as a result of role assignment us-
ing the two algorithms. This is essential as SBS algo-
rithm finds an optimal set of broadcaster so consumes
minimum energy. Our aim is to check whether the
new approach is efficient enough to assign roles such
that it does not consume too much of energy. In this
experiment also, we added new participants equiva-
lent 5% of the data set size. Fig. 6 shows system en-
ergy consumed when roles are assigned by SBS and
Crowdsourcing Location Sensitive Data for Dynamic Scenario by Adaptive Role Assignment
51

Figure 5: Time taken by SBS and Incremental Insertion Al-
gorithms for role assignment.
incremental insertion approach. It can be observed
that energy consumed by the proposed algorithm is
almost equivalent to that of the optimal algorithm.
Figure 6: Energy Consumption by SBS and Incremental In-
sertion Algorithms.
Next, we evaluate the nature of processing time
on a dataset of size 300 as we incrementally insert 15
participants. From Fig. 7, we observe that the rate of
processing remains same for every insertion in pro-
posed algorithm. However, there is slight variation
for SBS algorithm which happens when an inserted
node increases or decreases the number of broadcast-
ers. We also observe that time taken by the SBS al-
gorithm steadily increases. This is because the size of
the dataset increases with every insertion.
In the last set of experiment, we analyse a thresh-
old after which the SBS algorithm must be repeated.
This is evaluated by varying percentage of insertion
and observing the difference in energy between the
incremental and SBS algorithms. For this experiment,
we consider dataset of size 100, and incrementally in-
sert 5%, 10%, 15% and 20% nodes. From Fig. 8,
we can observe that when the number of insertions
is greater than 10%, the incremental approach con-
sumes more energy. So, 10% to 15% can be selected
as a threshold for rerunning SBS algorithm.
Figure 7: Time versus Number of Insertions.
Figure 8: Energy Consumption for {5%, 10%, 15%, 20%}
Insertions.
5.2 Incremental Deletion
In each of the results, we call the proposed algorithm
as Incremental Deletion and SBS for rerunning SBS
algorithm. Each of the following results was obtained
as an average of 25 runs, each of the size {100, 300,
500, 600}. For this, we considered one data set for
each of the sizes mentioned above and generated 25
sets of uniformly distributed random numbers. This
provided indices of participants which were used to
delete nodes.
In the first experiment, we evaluate time taken by
proposed incremental deletion and SBS algorithm for
role assignment. The number of participants which
were deleted was equivalent to 5% of the dataset size.
The sorting based algorithm had to be rerun for ev-
ery deletion. Fig. 9 clearly depicts that the proposed
algorithm takes much less time than SBS.
Next, we evaluate the impact on energy consump-
tion as a result of role assignment using the two al-
gorithms. This is required to evaluate performance
of proposed algorithm over the optimal algorithm as
energy is vital in mobile sensing. In this experiment,
WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems
52

Figure 9: Time taken by SBS and Incremental Deletion Al-
gorithms for role assignment.
we deleted 5% of the total participants in considera-
tion. Fig. 10 shows system energy consumed when
roles are assigned by SBS and incremental deletion
approach. It can be observed that energy consumed
by the proposed algorithm is almost equivalent to that
of the optimal algorithm.
Figure 10: Energy Consumption by SBS and Incremental
Deletion Algorithms.
In this experiment, we evaluate the nature of pro-
cessing time on a dataset of size 300 as we incremen-
tally delete 15 participants. From Fig. 11, we ob-
serve that the rate of processing remains same for ev-
ery deletion in the proposed algorithm. However, the
variation in SBS algorithm occurs due to the impact
of deletion on number of broadcasters. We also ob-
serve that in contrast to insertion, the time taken by
SBS algorithm steadily decreases. This is because the
size of the dataset decreases with every deletion.
In the last set of experiment, we analyse a thresh-
old after which the SBS algorithm must be repeated.
This is evaluated by varying percentage of deletion
and observing the difference in energy between pro-
posed and optimal algorithm. For this experiment, we
consider dataset of size 100, and incrementally delete
Figure 11: Time versus Number of Deletion.
5%, 10%, 15% and 20% nodes. From Fig. 12, we can
observe that when the number of deletions is greater
than 5%, incremental approach consumes more en-
ergy. So, a threshold between 10-15% can be chosen
for rerunning the SBS algorithm.
Figure 12: Energy Consumption for {5%, 10%, 15%, 20%}
Deletions.
6 CONCLUSIONS
The existing techniques based on greedy and sort-
ing based approaches are not efficient to consider the
adaptive changes to the dataset for role assignment.
Only way to assign role to new participant or change
roles of existing devices when some devices leave the
region is to rerun the algorithm. But as the number
of participants can be quite large, it is not worth re-
running the algorithm for each change. In lieu of this
concern, we propose incremental algorithms that help
in saving 95-99.9% of the time for role assignment
compared to the existing approach. Experimental re-
sults validate effectiveness of the proposed solutions.
Also when change of the dataset is within [5%, 10%],
the role assignment using these algorithms provide al-
most same energy consumption as obtained with the
optimal algorithms. The current state of work does
Crowdsourcing Location Sensitive Data for Dynamic Scenario by Adaptive Role Assignment
53

not consider velocity and movement information of
participants for the role assignment. It is essentially
based on their location registered to the server. This
is part of our future work.
An important aspect to receive good user partici-
pation is privacy. If security is endangered then de-
vices refrain from sharing their data. We plan to ex-
tend the work to incorporate several confidentiality
and privacy concerns along with incentive mechanism
to collect sufficient data samples.
ACKNOWLEDGEMENTS
This work was supported by the Natural Sciences and
Engineering Research Council of Canada (NSERC)
under grant number CRDPJ 476659-14.
REFERENCES
Anubhuti Garg, A., Nayak, A., 2017. Effective Role-
Assignment for Participatory Sensing Systems, In
IEEE Percom Workshop on Crowd Assisted Sensing,
Pervasive Systems and Communications.
Wang Wang, W., Xi, T., Ngai, E.C, Song, Z., 2016. Energy-
Efficient Collaborative Outdoor Localization for Par-
ticipatory Sensing, In Sensors, vol. 16, pp. 1–12.
Survey Lane, N.D., Miluzzo, E., Lu, H., Peebles, D.,
Choudhury, T., Campbell, A.T., 2010. A Survey of
Mobile Phone Sensing, In IEEE Comm. Magazine,
vol. 48, issue 9, pp. 140–150.
Sensors E. Macias, A. Suarez, J. Lloret, “Mobile Sensing
Systems”, Sensors, vol. 13, pp. 17292–17321, 2013
Sayler I. Constandache, S. Gaonkar, M. Sayler,
R.R. Choudhury, L. Cox, “Enloc: Energy-efficient
localization for mobile phones”, Proc. of IEEE
INFOCOM, pp. 2716–2720, 2009.
Yoon C. Yoon, D. Kim, W. Jung, C. Kang, H. Cha, “App-
Scope: Application Energy Metering Framework for
Android Smartphone Using Kernel Activity Monitor-
ing”, Proc. of USENIX ATC, pp. 387-400, 2012.
Geolife Y. Zheng, X. Xi, W.Y. Ma, “GeoLife: A Collabora-
tive Social Networking Service among User, Location
and Trajectory”, IEEE Data Eng. Bull. 33, pp. 32–39,
2010.
Song Z. Song, J. Ma, M. Dong, W. Wang, X. Gong,
X. Que, “Phoneradar: Infrastructure-free device-to-
device localization”, Proc. of IEEE Vehicular Tech.
Conf., 2014.
Kumar S. Kumar, V. Sharan, R. M. Hegde, “Energy efficient
optimal node-source localization using mobile beacon
in ad-hoc sensor networks”, Proc. of IEEE GLOBE-
COM, pp. 487-492, 2013.
Xiao B. Xiao, H. Chen, S. Zhou, “Distributed Localiza-
tion Using a Moving Beacon in Wireless Sensor Net-
works”, IEEE Trans. on Parallel and Distributed Sys-
tems, vol. 19, no. 5, pp. 587–600, 2008.
Davies N. Davies, A. Friday, P. Newman, S. Rutlidge,
O. Storz, “Using bluetooth device names to support
interaction in smart environments”, Proc. of ACM Mo-
biSys, pp. 151-164, 2009.
Johnson T.A. Johnson, P. Seeling, “Localization Using
Bluetooth Device Names”, MobiHoc, 2012.
Huang A. Huang, L. Rudolph, “A privacy conscious blue-
tooth infrastructure for location aware computing”,
Proc. of SMA Symposium, 2005.
WLAN I. Shafer, M.L. Chang, “Movement Detection for
Power-Efficient Smartphone WLAN Localization”,
Proc. of MSWiM, pp. 81-90, 2010.
QoI Z. Song, B. Zhang, C.H. Liu, W. Wang, “QoI-
Aware Energy-Efficient Participant Selection”, Proc.
of SECON, pp. 248-256, 2014.
Jigsaw H. Lu, J. Yang, Z. Liu, N.D. Lane, T. Choudhury,
T. Andrew, “The Jigsaw continuous sensing engine for
mobile phone applications”, SenSys, pp. 71-84, 2010.
Goldman J. Goldman, K. Shilton, J. Burke, et al.
“Participatory sensing a citizenpowered ap-
proach to illuminating the patterns that shape
our world.” http://www.wilsoncenter.org/sites/def
ault/f iles/participatorysensing.pdf
Hicks J. Hicks, N. Ramanathan, D. Kim, M. Monibi, J. Sel-
sky, M. Hansen, D. Estrin “AndWellness: An Open
Mobile System for Activity and Experience Sam-
pling.” Proceedings of the 1st Wireless Health Con-
ference, pp. 34–43, 2010.
Reddy S. Reddy, K. Shilton, G. Denisov, C. Cenizal, D. Es-
trin and M. Srivastava “Biketastic:sensing and map-
ping for better biking.” Proceedings of the 28th An-
nual CHI Conference on Human Factors in Comput-
ing Systems pp. 1817-1820, 2010.
PSense P. Baier, F. Durr, K. Rothermel “PSense: Reduc-
ing Energy Consumption in Public Sensing Systems.”
Advanced Information Networking and Applications,
pages 136-143, 2012.
WINSYS 2017 - 14th International Conference on Wireless Networks and Mobile Systems
54