This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2751304, IEEE Access
Practical Incentive Mechanisms for IoT-based
Mobile Crowdsensing Systems
Zhuojun Duan, Ling Tian , Mingyuan Yan, Zhipeng Cai, Qilong Han, and Guisheng Yin
AbstractThe powerful smart terminals with rich set of
embedded sensors promote the development of the Internet of
Things (IoTs). Mobile Crowdsensing system can be formed by
these mobile smart terminals from IoTs to collect and exchange
data. The main idea of MCSs is to outsource sensing tasks
(collecting data) to various mobile devices which are carried
by people or vehicles. The design of incentive mechanisms in
MCSs is one of the hottest research topics. However, most of the
existing literatures focus on maximizing the utilities or social
welfare while neglecting the practical requirements of MCSs
surveillance applications. In this paper, we discuss the importance
of fairness and unconsciousness of MCS surveillance applications.
Then, we propose offline and online incentive mechanisms with
fair task scheduling based on the proportional share allocation
rules. Furthermore, to have more sensing tasks done over
time dimension, we relax the truthfulness and unconsciousness
property requirements and design a (, )-unconsciousness online
incentive mechanism. Real map data are used to validate these
proposed incentive mechanisms through extensive simulations.
Index TermsOnline auction, incentive mechanism, Internet
of Things (IoTs)
I. I NTRODUCTION
The highly distributed paradigm Internet of Thing (IoT)
extends ubiquity of the Internet through integrating every
terminal for interaction via embedded systems, in which all
the physical terminals can collect and exchange data [1], [2],
[3], [4], [5]. IoT will be the fast-growing, largest market
potential and the most attractive emerging economy according
to the Top 10 Predictions of 2014 by Gartner [6]. In IoT, the
new emerging techniques integrate multiple types of sensors
and high-performance processors into physical terminals, e.g.,
smartphones (iPhones, Sumsung Galaxy, etc.), tablets (iPad,
etc.), and vehicle-embedded sensing terminals (GPS). These
mobile terminals can be used to sense and collect data, so
that become data sources. All above mentioned properties
make IoT a perfect choice for the Mobile Crowdsening System
(MCS). In an MCS, a complicated sensing job is divided
into several simpler tasks. Each participated mobile physical
Zhuojun Duan and Zhipeng Cai are with the Department of Computer
Science, Georgia State University, Atlanta 30303, GA, USA. E-mail: zdu-
an2@student.gsu.edu; zcai@gsu.edu.
Ling Tian is College of Computer Science and Engineering, University of
Electronic Science and Technology of China Chengdu, P.R.China 611731.
E-mail: lingtian@uestc.edu.cn.
Mingyuan Yan is Department of Computer Science and Information Sys-
tems, University of North Georgia, 82 College Cir, Dahlonega 30597, GA,
USA. E-mail: mingyuan.yan@ung.edu.
Qilong Han and Guisheng Yin is with the College of Computer Science and
Technology, Harbin Engineering University, 145 Nan Tong Da Jie Nangang,
Haerbin 150001, Heilongjiang, China. E-mail: hanqilong@hrbeu.edu.cn; yin-
guisheng@hrbeu.edu.cn.
To whom correspondence should be addressed.
2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
terminal can undertake one or more simpler sensing tasks. The
most attractive properties of MCSs is that it aims at letting the
regular mobile physical terminals work for the complicated
job, while keeping the users of these mobile physical terminals
unconscious. In tradition, however, the job must be done by
professional experts and the sensors have to be deployed in
advance.
The MCSs have already been applied to our daily life. One
of the best examples is the smart city project discussed in [7].
It can be used to collect information around the city and then
contributes to the intelligent operation of public services. In
detail, it tracks public vehicles and map bumps on the road
for the urban transportation systems in a city. The Microblogs
presented in [8], provides a mechanism where mobile physical
terminals can share their information (like travel, restaurants,
and news) through a universal platform. Then, the center server
in the platform processes and analyzes the shared data and
provides an alternative solution for problems or helps to make
decisions. MCSs can also be used in surveillance applications,
such as monitoring pollution levels or traffic, measuring water
levels, and collecting wildlife habitats. Practical surveillance
applications include Common Sense and CreekWatch which
can be found in [9] and [10], respectively.
However, users of these mobile physical terminals partici-
pating in an MCS will suffer from extra resource consumption
(battery and computing capacities) and the risk of privacy
exposure (location exposure). So effective and efficient in-
centive mechanisms are needed in MCSs to attract enough
mobile physical terminals participation. A common strategy
designed in MCSs is to give rewards to participated users as
compensation and stimulation. Lots of works can be found on
incentive mechanisms and most of them are based on game
theory [11], [12]. We classify the existing works into two
categories: the offline incentive mechanisms and the online
incentive mechanisms. The former will collect the information
of all participants before making the decision, while the
latter decides whether to accept a new arriving participant
sequentially without the information of next following partic-
ipants. After analyzing these existing works, we find they are
not appropriate to surveillance applications for the following
reasons: i) the tasks allocation algorithms are unfair over the
time dimension. Most of the surveillance applications require
continuous sensing information for a period of time. Taking
the noise level monitoring application as an example, if a cloud
center wants to surveil the noise level of a place, it expects
to get noise data of the place for a period of time. Gener-
ally speaking, more than one mobile physical terminal will
participate in the sensing task. It is better to evenly schedule
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2751304, IEEE Access
sensing tasks among a set of mobile physical terminals over
the particular period of time. However, mechanisms proposed
in existing works [9][13] may lead to the situation that several
mobile physical objects are assigned to sense the noise at
the same time incidentally. ii) existing mechanisms require
deep interaction between participated users and their mobile
physical terminals. That is, sometimes participated users are
required to pay lots of attention to their devices or forced to
change their own schedule when working for the MCS. For
example, the working schedules of users are decided by the
MCS in [14]. The significant advantage of MCSs over Wireless
Senor Networks (WSNs) is that we dont need to deploy
the sensors or workers in advance. However, the required
deep interaction will interfere participants original plan which
should be avoided. The incentive mechanisms investigated in
this work try to overcome the two weaknesses.
We consider an MCS on surveillance applications from both
the time and space dimensions. Each sensing task published
by the MCS is tagged with a location requirement and a period
of time requirement. Once being published, the sensing task
is required to be done multiple times over the time period at
the specific location. For fairness over the time dimension, a
time period is divided into smaller time slots and the sensing
task will be processed periodically over these time slots. The
participants of the MCS are the mobile vehicles with sensors
installed and are able to work for sensing tasks. The device
on the vehicle will communicate with cloud servers by 3G or
LTE techniques. The drivers of these vehicles are general office
workers commuting between home and office. Their routes are
relatively stable and they will let the servers know their routes
in advance. When a vehicle passes through a location where a
sensing task is required, the sensors can work for the sensing
task automatically. The objective of the MCS is to select a set
of qualified vehicles with devices so that as many tasks can
be done evenly over time.
In this paper, we first design an offline incentive mechanism
where the proportional share allocation rule is applied. Then
we consider the realistic situations and propose online auction
mechanisms where each winner vehicle will be decided relying
on the information of the vehicle itself and the vehicles
arriving before it. The contributions of this paper are as
follows:
We first discuss and investigate the importance of un-
consciousness in MCSs and get the conclusion that the
frequencies of interaction between participants and cloud
center should be minimized.
We introduce the MCS model on surveillance applica-
tions. After that, the design of incentive mechanisms
under the offline and online cases are designed. The task
allocation algorithms are implemented fairly considering
practical property requirements.
In order to improve the performance of the online
incentive mechanism, we relax the truthfulness and
unconsciousness requirements and propose a (, )-
unconsciousness online incentive mechanism.
In the rest of the paper, we present and discuss previous
works in section II. Then the MCS system model and problem
2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
are formulated in section III. Incentive mechanisms for the
offline and online cases are introduced in section IV and
section V, respectively. We evaluate the performance of these
proposed incentive mechanisms in section VI and conclude
the paper in section VII.
II. R ELATED WORKS
In this section, existing literature related to our research
are summarized. We first present the existing research on
commercial MCS applications. After that, the advantages and
disadvantages of existing research on incentive mechanisms
in MCSs are discussed. Finally, existing online incentive
mechanisms in MCSs are presented.
Commercial MCS applications can be divided into three
categories[15]. The first category is environmental applica-
tions, where participants undertake tasks for environmental
monitoring, such as pollution level measuring, noise level
monitoring and wildlife habitats surveillance. Example ap-
plications include Common Sense[9] and CreekWatch[10].
The second category focuses on infrastructures. Applications
in this category collect information from participated users
and then use the information to help the decision making on
improving public services (like parking availability and traffic
conditions). For example, Smart Cities introduced in [7] track
the urban environment and use the collected information to
help improve the quality of life in or around cities. CarTel
and Nericell, presented in [16] and [17] respectively, are also
infrastructure applications. The last category contains social
applications where participants can sense their own data and
share them through social networks. A few representative
works can be found in DietSense[18] and BikeNet[19]. All
these applications mentioned above demonstrate the great
potential of MCSs.
Now, many researchers focus on incentive mechanisms
design in MCSs and most of which are based on game
theory[20][21][22][23]. Two types of incentive mechanisms
are considered in [24]: the platform-centric incentive mecha-
nism and the user-centric incentive mechanism. The former
is based on Stackelberg game where the platform controls
the total budget distribution, and users adjust their strategies
to satisfy the requirements requested by the platform. In the
latter, users bid with a price for the services they provided
to the platform. But [24] only considers the offline case. In
[25], the authors focus on the problem of distributed time-
sensitive and location-dependent task selection. The problem is
formulated as a non-cooperative task selection game (TSG). A
distributed solution is proposed based on the potential function
[26] in game theory. However, the work scheduling algorithm
proposed in [25] only considers the utility of each user. So
working for sensing tasks may disturb the normal lives of
participants. As we discussed in the last section, an attractive
incentive mechanism should keep its influence on participants
normal life as low as possible.
Literature [27][28][14] study the design of online incen-
tive mechanisms design. [27] investigates the online task
assignment problem and provide a two-phase exploration-
exploitation assignment algorithm. Authors of [28] formu-
late the task assignment and label inference problems for
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2751304, IEEE Access
Base Station
Crowdsourcing
Platform
Bid
Payment
Base Station
$$$
Mobile Vehicle with Device
Sensing task
Fig. 1. An overview of a Mobile Crowdsensing System (MCS).
heterogeneous classification tasks. They adopt online primal-
dual techniques to design the probably near-optimal adaptive
assignment algorithm. But [27] and [28] omit discussing the
reward for participants. [14] focuses on the online incentive
mechanism design and proposed OMZ for the zero arrival-
departure interval case and OMG for the general case. Howev-
er, the task allocation discussed in this paper does not consider
the time-dimension. There is a potential that a task is allocated
to several users but these users process the task at the same
time. That will bring waste of resources in MCS surveillance
applications.
III. S YSTEM MODEL AND PROBLEM FORMULATION
A. Problem Formulation
Considering a Mobile Crowdsensing System (MCS) (as
shown in Fig. 1) which is able to undertake sensing tasks such
like traffic surveillance and environmental pollution monitor-
ing. In the MCS, a Crowdsensing Platform (CP) publishes a
set of sensing tasks = {1 , 2 , ..., m } (|| = m). Each
sensing task is defined by a collection of features:
= (l , r ),
where l L specifies the location of the sensing task and
L represents the set of locations which are along routes. r
represents the reward that the CP would like to pay if the
sensing task is done. Each sensing task in is required to be
sensed during time T . For simplicity, T is divided into multiple
time slots T = {t1 , t2 , ..., t|T | }. Let V = {v1 , v2 , ..., vn }
be the set of Mobile Vehicles with Devices (MVDs). These
mobile vehicles move on routes and will pass through one
or more locations of sensing tasks. Corresponding sensing
tasks will be performed by the devices installed on MVDs
automatically. Given an MVD v V , its features can be
denoted as
v = {v , cv , v },
where v is the route of v and is defined as
discrete location-time points information v
{(lva , tav ), (lv1 , t1v ), ..., (lvi , tiv ), ..., (lvd , tdv )}. Each element
(lvi , tiv ) in v indicates v will pass through location lv at time
slot tiv by estimation. (lva , tav ) and (lvd , tdv ) are used to represent
vs starting and destination location-time points, respectively.
2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Assume a vehicle would not visit a location more than once
in T . An MVD is able to finish any sensing task if the MVD
passes through the location of the sensing task. Let cv be
the cost of v if v works for all the sensing tasks located
in its route. v is vs driving speed which determines how
many time slots are required for v to move between any two
different sensing tasks.
All these surveillance sensing tasks (traffic surveillance
or environmental pollution monitoring) require to be sensed
multiple times in T. However, it is difficult to persuade an
MVD to stay at a location without influencing its original
routine. Alternatively, a sensing task could be sensed multiple
times by different MVDs over different time slots. We call the
number of times a sensing task required to be sensed as its
space-time coverage requirement. To be fair, for each sensing
task, its space-time coverage requirement is distributed over
the time slots in T evenly. Matrix F = [f,t ] (0, 1)T is
used to represent the space-time coverage requirements of all
sensing tasks over the time dimension. For example, f,t = 1
represents that the sensing task needs to be sensed once
in time slot t. Otherwise, f,t = 0. The objective of the CP
is to choose winner MVDs, set W that can reach the best
coverage requirements over all sensing tasks. The problem can
be defined as,
P P P
M aximize f,t xv,,t
vW tT
s.t. xv,,t {0, 1}
(1)
P
xv,,t f,t , , t T
vV
W V
where W is the winner MVDs. Matrix xv = [xv,,t ]
(0, 1)T represents the allocated working schedule for v.
xv,,t = 1 indicates v is allocated to work for sensing task
in time slot t, otherwise xv,,t = 0. The second constraint
specifies that a sensing task should be allocated to no more
than one MVD in a specific time slot.
B. Reverse Auction Model Design
Working for sensing tasks brings extra battery consumption,
hardware loss and privacy threats to MVDs. Therefore, the
winner MVDs expect to receive monetary rewards from the
CP as stimulation and compensation. We apply reverse auction
model to the interaction between the CP and MVDs, where
the CP acts as the buyer and auctioneer at the same time. The
roles of MVDs in the model are sellers.
After the CP publishes the sensing tasks, each v V
submits its bid, which can be denoted as,
bv = {v , Av }
= where v is the set of location-time points that v will pass
through. Av is the asking price when v is selected as a winner
to work for these sensing tasks on its route. If the reverse
auction mechanism is truthful, v = v and Av = cv . That
is all MVDs will submit their real route and take the asking
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2751304, IEEE Access
price as their base price. Assume, all MVDs are single-minded
(Definition 1) so that they have simple cost functions.
Definition 1. A cost function c() is called single-minded if
there exists a sensing tasks allocation S and a cost c
such that c(S) = c for any S S and c(S) = for
all other S.
For each MVD v V in our model, S = {(, t)|(, t)
v } is the set of sensing tasks can be done by v and S denotes
the sensing tasks allocated to v by the CP when v wins in the
auction. Therefore, once a winner v is allocated any set of
sensing tasks which v is able to sense, its cost is a consent
value. If the allocated sensing task set includes one or more
sensing tasks which v can not sense, v will reject the allocation
and the cost of v is set as infinity for clarity. Each v sets its bid
according to the strategy aiming to maximize its own utility.
Uv is used to denote the utility of v and defined as:
pv cv vv wins,
Uv =
0 otherwise.
Generally speaking, the incentive mechanism should satisfy
several properties to guarantee its efficiency and effectiveness.
Individual Rationality. Because all MVDs are self-interest
to benefit themselves, the utility of any v V should be non-
negative: Uv 0.
Truthfulness. An auction mechanism is called truthful if all
MVDs bid with their true value (real cost). The utility of vj
will be maximized when it reports true values in its bid and
vj cannot improve its utility through any misreport:
Uvj (bvj , bvj ) Uvj (bvj , bvj ),
where, bvj = {bv1 , ... , bvj1 , bvj+1 , ... , bvn } represents
the set of truthful bids of all MVDs excluding vj . bvj is the
truthful bid of vvj , and bvj 6= bvj . If an auction mechanism
satisfies this property, N ash Equilibrium exists.The misre-
ports of first value (route) in a bid can be easily detected by
the CP through the submitted results of their works. Thus the
truthfulness of the first value is guaranteed. We focus on the
truthfulness of the second value in a bid: asking price.
Budget Balance. The upper bound
P of the total payments
for all the MVD winners is B= r , and we call B as
the budget constraint of the CP. In other word,
P the auction
mechanism should be budget balance: B pv .
Unconsciousness. Participation for the MCS are subordi-
nate to MVDs original target. In detail, the route of each
MVD has been scheduled before the CP publishes the sensing
tasks. An MVD will not change its route for the reward. On
the other hand, when an MVD passes through the location of
a sensing task, the sensors installed on the MVD should work
automatically without requiring operation from the driver. We
call this kind of participation as unconsciousness.
Computational Efficiency. An auction mechanism is con-
sidered computationally efficient if the task allocation and
payment decision can be implemented in polynomial time.
When the above properties are all satisfied, an auction
mechanism can be considered as useful. Without individual
2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
T
Publication Auction Working t
Stage Stage Stage
Fig. 2. Offline interaction process between the CP and MVDs.
rationality, an MVD may receive negative utility, and refuses
to participate in the MCS. Then, because the cv in bid bv
is private to v, the CP wouldnt know it. If an auction
mechanism is truthful, all MVDs just need to bid with their
true costs: Ai = ci , which not only simplify the strategies, but
also avoid possible manipulation from some MVDs. Budget
balance make all winner MVDs get their deserved payments.
Unconsciousness attracts more MVDs to participate in the
MCS. Finally, computational efficiency will guarantee that the
auction mechanism can be practically implemented.
(2)
IV. O FFLINE AUCTION MECHANISM
A. Offline Working Process of MCSs
In this section, we first focus on the design of offline
incentive mechanisms. The working process of an offline MCS
can be divided into three stages: publishing stage, auction stage
and working stage, as shown in Fig. 2.
Publishing stage. In this stage, the CP decides the sensing
tasks that it plans to finish within T . Then it publishes the
description of these sensing tasks among the MVDs.
Auction stage. After receiving requirements of sensing
(3) tasks and their description, each MVD generates location-time
points sequence according to its original scheduled route. The
sequence of location-time points implies the set of sensing
tasks an MVD can take. If an MVD is able to work for a
set of sensing task , it will further evaluate the cost caused
by them. An MVD calculates its cost as the base price and
submits a bid to the CP. The bid submitted by an MVD consists
of its location-time points sequence and the base price. After
receiving bids from all participating MVDs, the CP will choose
a set of winner, make the work schedule, determine each
winners reward and then announce the auction result to all
participated MVDs.
Working stage. According to the working schedules, each
MVD winner will be activated by the CP while passing
vW through a specific location at a specific time. The reward is
given to an MVD once it finishes all allocated sensing tasks.
In this work, our focus is the design of efficient and effective
incentive mechanisms during the auction stage. The other two
stages are omitted.
B. Modified Proportional Share Auction Mechanism
The design of an offline incentive mechanism for problem
(1) is more complex than our past work [29] because the
consideration of the budget balance property. We rewrite the
problem function in (1) as a new form g(W ) = | Sv |,
vW
where Sv = {(, t)|(, t v , f,t = 1)} and find an
interesting point: it is a nondecreasing submodular function.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2751304, IEEE Access

Definition 2. A function h() is submodular if:

h( {v}) h() h(X {v}) h(X), t
Auction Stage
where is a finite set, X and v \X, and Publication
Stage
&
h() : 2V R+ . Working Stage

Theorem 1. The objective function g(W ) is a nondecreasing

submodular function.
Fig. 3. Online interaction process between the CP and MVDs.
Proof. For any W X V and v V \X, there have
Sv0 0 Sv0 and Sv ( 0 Sv0 ) Sv ( 0 Sv0 ),
v 0 W v X v W v X
so we can get coverage requirement until condition Av gg(W v (W )B
v) becomes
g(W v) g(W ) = |Sv | |Sv ( 0 Sv0 )| false.
v W Once the winner set is identified, payment of each winner v
|Sv | |Sv ( 0 Sv0 )| in W will be calculated as follows. Firstly, sort all vj W \v
v X
= g(X v) g(X). in the non-increasing sorting as,
Then,it is easy to obtain a conclusion gvv (O0 ) g v (O1 ) gvv (Oj1 ) gvv (On2 )
1
v2 ... j ... n1 ,
g(X) g(W ) = |( Sv0 ) ( Sv )| 0, Av1 Av2 Avj Avn1
v 0 X\W vW (4)
so g(W ) is nondecreasing. where Oj represents the set of first j MVDs in the sorting
result (O0 = ) and gvv j
(Oj1 ) is the marginal contribution
Based on the above analysis, we apply the modified pro- of vj when v is removed. Then find the MVD v 0 W \v
portional share auction mechanism proposed in [30], which is in the position z of the sorting result which satisfies Avz
based on the proportional share allocation rule. The auction gvv (Oz1 )B
z
g(Oz ) . The payment of v will be determined by,
mechanism has two stages: winner set determination and
payment decision. v
gv(j) (Oj1 )Avj v
gv(j) B
pv = max {min{ , }},
Algorithm 1 Winner set determination j[1,2,...,z+1] gvj (Oj1 ) g(Oj1 {v})
(5)
Input: v
where gv(j) (Oj1 ) = g(Oj1 {v})g(Oj1 ) represents the
{B; F ; (bv , v V )} marginal contribution of v at position j in the sorting result.
Output: Theorem 2. The modified proportional share auction mech-
{W ; pv and xv , v W } anism satisfies: individual rationality, truthfulness, Budget
Balance, and computational efficiency [30].
1: Initialization:
Theorem 3. Participation in the MCS are unconscious to all
2: W = , v arg maxv0 V (gv0 (W )/Av0 ).
MVDs.
3: xv,,t = 0, v V , , t T ;
Proof. The working scheduling for each winner MVD is on its
4: y,t = 0, , t T . predefined route and the MVD will be triggered automatically,
5: while Av gg(W
v (W )B
v) do
so theorem 3 is true.
6: W = W v;
7: v arg maxv0 V \W (gv0 (W )/Av0 ). V. O NLINE REVERSE AUCTION MECHANISM
v v
8: for each (l , t ) v , each do A. Online Working Process of MCSs
9: if l is same to lv and y,tv == 0 then
10: xv,,tv = 1, y,tv = 1. In this section, we try to solve the problem formulated
in section III online. Compared with the offline interactions
11: end if process in MCSs, the online interaction between the CP and
12: end for MVDs are more flexible. The auction stage and working stage
13: end while are mixed, as shown in Fig. 3. The CP will publish the sensing
tasks in advance. Then for any MVD, it can participate in the
The winner set determination process is shown in algorithm MCS and submit its bid at anytime within T . Once the CP
1, where gv (W ) denotes the marginal contribution of v to the receives the bid, it will immediately determine whether the
coverage requirements, and is calculated as: MVD wins or not. If the MVD wins, the CP will make the
working schedule and determine the payment for this MVD.
gv (W ) = g(W v) g(W ).
Then the MVD will work for the sensing tasks according to the
The winner set determination algorithm iteratively selects received working schedule. After all scheduled sensing tasks
the MVD who has the largest marginal contribution to the are done, the CP will make payment to the MVD.

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2751304, IEEE Access

B. Simple Online Incentive Mechanism Computational efficiency: the number of location-time

points in v is bounded by m. So the time complexity of
When design online incentive mechanisms, one precondition
algorithm 2 is O(m m).
should be kept in mind: the CP has no knowledge about the
upcoming MVDs and isnt able to make predictions about Theorem 5. An incentive mechanism is truthful if and only if
that. In order to satisfy these property requirements discussed it is monotone and the payment for each winner is a critical
in section III, we first propose a simple online incentive value [20].
mechanism, which is also based on the proportional share
allocation rule as shown in algorithm 2. For each new coming
C. (, )-unconsciousness Online Incentive Mechanism
v, we first calculates a temporary payment p0v for v which is
proportional to the marginal contribution of v over all coverage Simple-OIM is simple and able to determine the winner
requirements (line (1)). If p0v isnt smaller than the asking price MVD set and make payment decision. In order to further
Av , v will win. Its payment pv = p0v and its working schedule improve the performance, a new online incentive mechanism
will be set (line (5-6)). is proposed which targets at covering more sensing tasks over
time with relaxed truthfulness and unconsciousness require-
Algorithm 2 Simple Online Incentive Mechanism (Simple ments. Our new online incentive mechanism is motivated by
OIM ) the following two facts. First, most of the incentive mecha-
Input: {B; F ; W ; bv ; Y } nisms achieve truthfulness at the expense of effectiveness. Our
objective is to get as many sensing tasks covered over time as
Output: possible within a limited budget. Based on this concern, the
{W ; pv ; xv ; Y ; B} real cost of each MVD is not crucial to the CP. Second, the
gv (W ) most sensitive information of a route are the source location,
1: p0v = g(F) B; pv = 0;
destination location, and the total time duration. Taking a
2: if p0v
Av then commuter as an example. Most of the time, a commuter will
3: W = W v, pv = p0v ; drive from home in the morning. He or she should arrive
4: for each (lv , tv ) v , each do at office within a specific time duration. With a reasonable
5: if l is same to lv and y,tv == 0 then reward, the commuter probably accepts to take a new route
6: xv,,tv = 1, y,tv = 1, B = B pv ; from home to the office if the commuting time isnt extended
strongly. Thus, the idea of the new online incentive mechanism
7: end if is to recommend another alternative candidate route for losing
8: end for MVDs based on Simple-OIM. The candidate route should
9: end if maintain the lowest influence on the participating MVDs
unconsciousness. One step further, the candidate route should
As shown in theorem 4, the simple auction mechanism be as close to the original route of the commuter as possible.
presented in algorithm 2 satisfies all the desired auction In this way, more sensing tasks are expected to be covered
mechanism properties proposed in section III. by MVDs over time and the utility of the losing MVD can
also be increased if it accepts the recommended candidate
Theorem 4. Simple-OIM satisfies the desired individual ra- route. So the decision of candidate route is a trade-off between
tionality, truthfulness, budget balance, unconsciousness and utility and effectiveness. For simplicity, two new definitions
computational efficiency. are introduced here.
Proof. Individual rationality: an MVD becomes a winner only Definition 3. (, )-Potential Route ((,) ). The (,) of
under the condition that p0v Av , then there always has route ={(l0 , t0 ), (l1 , t1 ), ..., (l|| , t|| )} should satisfy:
Uv = pv Av 0 because pv = p0v . So individual rationality i) and (,) should start at the same location-time points
property is guaranteed; and end at the same destinations.
Truthfulness: an MVD v wins in the case that p0v = ii) the similarly degree between and (,) should be larger
gv (W ) 0
g(F) B Av . If v bids with Av Av , it will still win. than ( [0, 1]). The similarity degree is calculated as,
Thus we can say the incentive mechanism in algorithm 2 is
monotone. On the other hand, if Av p0v , v will win with |{l = l0 |l (l, t) and (l, t) , l0 (l0 , t0 ) and (l0 , t0 ) (,) }|
,
payment pv = p0v . v will lose otherwise. So p0v can be regarded |{l|l (l, t) and (l, t) }|
as the critical value for v. Therefore, according the theorem (6)
5, algorithm 2 is truthfulness; where l = l0 means that l and l0 are the same location.
Budget balance: for each winner v W , its payment is iii) the total travel time of (,) is no more than the total
calculated
P basedPongvline (1) of algorithm 2 and it is easy to travel time of plus a delay tolerance threshold ( 0),
(W )
get pv = g(F ) B B;
vW vW |(,) | ||
Unconsciousness: the route of each MVD doesnt change X X
( (tn tn1 )) (tn tn1 ) . (7)
due to its participation in the MCS and the mobile physical n=1, n=1,
objects of winner MVDs will be triggered automatically (ln ,tn )(,) (ln ,tn )
according to their working schedules;

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2751304, IEEE Access

(,)
Definition 4. Candidate Route (CR). represents the set
(,)
of potential routes for . Candidate route CR in is the
(, )-Potential Route of with largest marginal contributions,
that is CR arg max(,) (,) (g(,) (W )).

Based on the above discussion, we propose an online

incentive mechanism: (, )-OIM. For each v, let a symmetric
matrix Dv = [dvl,l0 ] [Z + ]LL represent the number of time
slots needed by v to travel between any two locations in L.
Specifically, values of Dv are based on the speed of v, denoted
by v . In the online incentive mechanism, each v submits its
bid in a new form:

bv = {v , Av , Dv }.

The detailed design of ( , )-OIM is shown in algorithm

3. It first applies the Simple-OIM for each new coming v (line
(1)). If v wins, vs payment and work schedule will be decided
by the Simple-OIM. Otherwise, a candidate route will be found
by tweaked Depth First Search (DFS) [31] and recommended
to v (line (3)). Here the tweaked DF S is a traversal algorithm
which can find all routes between two specific locations. If the Fig. 4. The locations distribution of tasks over Atlanta metropolitan area
candidate route exists, the payment and work schedule of v is (30km*40km).
decided as shown in line (6-12).
VI. P ERFORMANCE EVALUATION
Algorithm 3 (, )-unconsciousness Online Incentive Mech-
anism ((, ) OIM ) A. Evaluation of the offline case
Input: {B; F ; W ; bv ; Y ; ; } The sensing region is 30km*40km and located in the
Atlanta metropolitan area. We mark 22 popular locations
Output: within the region in Google map as sensing task locations
{W ; pv ; xv ; Y ; B} (shown in Fig. 4). Then the budget B varies from 2000 to
14000. The number of MVDs varies from 50 to 250. For
1: (B, W , pv , Y , xv )=Simple OIM (B, F , W , bv , Y );
each MVD, its speed and cost are randomly generated from
2: if v / W then
[25km/h, 60km/h] and [10, 30], respectively. The total time
3: Adopt the tweaked Depth F irst Search to find the
(T = 2.5h) is divided into 150 time slots. The route of each
candidate route CR v for v;
g(,) (W ) MVD is a sequence of locations on the map and the time at
4: if g(F ) B Av then when the MVD will pass through them is obtained based on
g(,) (W )
5: pv = B its speed. One step further, the starting time of the route is
g(F )
distributed over T .
6: W = W v; The total number of tasks covered by winner MVDs in the
7: for each (lv , tv ) CR
v , each do offline incentive mechanism is shown in Fig. 5. We observe
8: if l is same to lv and y,tv == 0 then that more tasks can be done with the increase of either the
9: xv,,tv = 1, y,tv = 1, B = B pv ; number of MVDs or the budget, respectively. Secondly, we
compare the coverage of sensing tasks in our proposed offline
10: end if
incentive mechanism with the mechanism designed in [14]
11: end for
which hasnt considered the time dimension. The results are
12: end if
shown in Fig. 6(a) and 6(b), separately. We can see the covered
13: end if
location-time points in Fig. 6(a) is denser than that in Fig. 6(b)
by about 20 percent. The result shows that the offline incentive
(, )-OIM gives MVDs who lose in the Simple-OIM mechanism in our paper gets more sensing tasks done over the
another opportunity to win. This leads to more MVDs working space and time dimensions. The reason for the sparse coverage
for the CP and more tasks can be done. in Fig. 6(b) is the overlapping: a sensing task may be covered
by more than one MVD at a specific time slot.
Theorem 6. (, )-OIM satisfies the individual rationality,
budget balance, and (, )-unconsciousness property require-
B. Evaluation of the online case
ments.
To evaluate the performance of online incentive mechanisms
Proof. The proof is similar to Theorem 4. designed in this paper, we take the secretary mechanism as a

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2751304, IEEE Access

2000
|B|=2000
1800 |B|=4000
|B|=6000
1600 |B|=8000
Coverage of locationtime points

|B|=10000
1400 |B|=12000

1200

1000

800

600

400

200

50 75 100 125 150 175 200 225 250

The number of MVDs

Fig. 5. The total number of tasks covered by the offline incentive mechanism. Fig. 7. The coverage of online incentive mechanisms

22

20

18

16

14
Task index

12

10

0
0 50 100 150
Time slots

(a)

22

20

18

16

14 Fig. 8. The average running time of online incentive mechanisms.

Task index

12

10

6
online incentive mechanisms proposed in this paper with the
4 secretary mechanism. The result is shown in Fig. 7. We can
2 observe that for each online incentive mechanism, its coverage
0
0 50
Time slots
100 150 increases with the increase of participating MVDs. Then, the
results of the secretary mechanism and the Simple-OIM are
(b)
almost same. However, first k MVDs is rejected in secretary
Fig. 6. (a)The location-time points coverage by offline incentive mechanism. mechanism which can not guarantee sovereignty because these
(b)The location-time points coverage by [14]. MVDs in K are excluded arbitrarily. The MCS should make
each MVD have the same opportunity to win. From this aspect,
Simple-OIM is better than the secretary mechanism. (, )-
benchmark which is based on the classical secretary algorithm OIM outperforms the other two mechanisms because it gives
[32]. That can be summarized as: the losing MVDs one more chance to win. We know that and
Secretary mechanism. Let the first k arrived MVDs as set K, are used to constrain the potential routes. Strict constraints
reject the MVDs in K, and calculate = max{ gvc(K) v
} as (larger of and smaller of ) will limit number of potential
vK
threshold. Then for each new coming MVD denoted as v 0 routes and decrease running time. But loose constraints will
which satisfies gvc0 (w)
0
, calculate its temporary payment lead to more tasks covered over time. Thus, (0.3, 20)-OIM
v
performs better than (0.7, 10)-OIM.
p0v0 = gv0(w) . v 0 will be selected as 0
P a winner (W = W v )
0
with payment pv0 = pv0 if pv0 + pv B. Then we test the performance of the four online incentive
vW mechanisms under different number of sensing tasks. The
The experimental setup of the online case is similar to the results of average running time for each MVD under different
offline case. The MVDs are required to submit bids at their setups are shown in Fig. 8. The average running time of each
starting time. We first compare the number of sensing tasks MVD in secretary mechanism and the Simple-OIM turns out
covered by winner MVDs over time obtained from the two to be negligible with the increase of the number of sensing

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2751304, IEEE Access
tasks. Because the tweaked DF S is adopted to calculate the
potential routes for each losing MVD, the time complexity
is high in (, )-OIM. So (0.3, 20)-OIM needs more time
than (0.7, 10)-OIM. In order to guarantee the computational
efficiency, the (, )-OIM should choose larger and smaller
even though the part of effectiveness will be sacrificed.
VII. C ONCLUSION
In this paper, we focus on incentive mechanisms design in
IoT-based MCSs for surveillance applications. We investigate
practical requirements and the importance of fairness and
unconsciousness in winner MVDs selection. Two kinds of
incentive mechanisms are proposed which can be applied in
realistic applications. Extensive simulations are conducted to
validate the performance of them.
ACKNOWLEDGMENT
This work is partly supported by the National Science
Foundation (NSF) under grant NOs 1252292, 1741277 and
1704287, the Foundation of Science & Technology Depart-
ment of Sichuan Province No. 2017JY0007, 2017HH0075,
2017JY0073, 2017JY0027, 2016GZ0075, 2016GZ0077, and
the NSFC under grants No. 61373083, 61370084, and
61502116.
R EFERENCES
[1] Z. Wang, X. Xu, K. Zhou, and H. Cai, Mobile crowd sensing: an
emerging paradigm for internet of things, International Journal of
Autonomous and Adaptive Communications Systems, vol. 10, no. 2, pp.
174191, 2017.
[2] T. Qiu, X. Liu, M. Han, M. Li, and Y. Zhang, Srts: A self-recoverable
time synchronization for sensor networks of healthcare iot, Computer
Networks, 2017.
[3] T. Qiu, A. Zhao, R. Ma, V. Chang, F. Liu, and Z. Fu, A task-efficient
sink node based on embedded multi-core soc for internet of things,
Future Generation Computer Systems, 2016.
[4] T. Qiu, R. Qiao, and D. Wu, Eabs: An event-aware backpressure
scheduling scheme for emergency internet of things, IEEE Transactions
on Mobile Computing, vol. PP, no. 99, pp. 11, 2017.
[5] X. Zheng, Z. Cai, J. Yu, C. Wang, and Y. Li, Follow but no track:
Privacy preserved profile publishing in cyber-physical social systems,
IEEE Internet of Things Journal, vol. PP, no. 99, pp. 11, 2017.
[6] Top 10 strategic technology trends for 2017: A gartner trend insight
report, http://www.gartner.com, accessed: 2017-07-30.
[7] r. Petkovics, V. Simon, I. Gdor, and B. Brcz, Crowdsensing solutions
in smart cities towards a networked society, vol. 1, 10 2015.
[8] S. Gaonkar, J. Li, R. R. Choudhury, L. Cox, and A. Schmidt, Micro-
blog: Sharing and querying content through mobile phones and social
participation, in Proceedings of the 6th International Conference on
Mobile Systems, Applications, and Services, New York, NY, USA, 2008,
pp. 174186.
[9] P. Dutta, P. M. Aoki, N. Kumar, A. Mainwaring, C. Myers, W. Willett,
and A. Woodruff, Common sense: Participatory urban sensing using
a network of handheld air quality monitors, in Proceedings of the 7th
ACM Conference on Embedded Networked Sensor Systems, New York,
NY, USA, 2009, pp. 349350.
[10] S. Kim, C. Robson, T. Zimmerman, J. Pierce, and E. M. Haber, Creek
watch: Pairing usefulness and usability for successful citizen science, in
Proceedings of the SIGCHI Conference on Human Factors in Computing
Systems, New York, NY, USA, 2011, pp. 21252134.
[11] Z. Duan, W. Li, and Z. Cai, Distributed auctions for task assignment
and scheduling in mobile crowdsensing systems, in IEEE ICDCS, June
2017, pp. 635644.
[12] J. L. Z. Cai, M. Yan, and Y. Li, Using crowdsourced data in location-
based social networks to explore influence maximization, in IEEE
INFOCOM, 2016, pp. 19.
2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
[13] H. Jin, L. Su, D. Chen, K. Nahrstedt, and J. Xu, Quality of information
aware incentive mechanisms for mobile crowd sensing systems, in
Proceedings of the 16th ACM International Symposium on Mobile Ad
Hoc Networking and Computing. ACM, 2015, pp. 167176.
[14] D. Zhao, X.-Y. Li, and H. Ma, How to crowdsource tasks truthfully
without sacrificing utility: Online incentive mechanisms with budget
constraint, in INFOCOM 2014, 2014, pp. 12131221.
[15] R. K. Ganti, F. Ye, and H. Lei, Mobile crowdsensing: current state and
future challenges, IEEE Communications Magazine, vol. 49, no. 11,
2011.
[16] B. Hull, V. Bychkovsky, Y. Zhang, K. Chen, M. Goraczko, A. Miu,
E. Shih, H. Balakrishnan, and S. Madden, Cartel: a distributed mobile
sensor computing system, in Proceedings of the 4th international
conference on Embedded networked sensor systems, 2006, pp. 125138.
[17] P. Mohan, V. N. Padmanabhan, and R. Ramjee, Nericell: rich mon-
itoring of road and traffic conditions using mobile smartphones, in
Proceedings of the 6th ACM conference on Embedded network sensor
systems. ACM, 2008, pp. 323336.
[18] J. Burke, D. Estrin, and M. Hansen, Image browsing, processing, and
clustering for participatory sensing: Lessons from a dietsense prototype,
Retrieved October, vol. 12, p. 2013, 2007.
[19] S. B. Eisenman, E. Miluzzo, N. D. Lane, R. A. Peterson, G.-S. Ahn,
and A. T. Campbell, Bikenet: A mobile sensing system for cyclist
experience mapping, ACM Transactions on Sensor Networks (TOSN),
vol. 6, no. 1, p. 6, 2009.
[20] N. Nisan, T. Roughgarden, E. Tardos, and V. V. Vazirani, Algorithmic
game theory. Cambridge University Press Cambridge, 2007, vol. 1.
[21] J. Sun and H. Ma, Heterogeneous-belief based incentive schemes for
crowd sensing in mobile social networks, Journal of Network and
Computer Applications, vol. 42, pp. 189196, 2014.
[22] Y. Wang, G. Yin, Z. Cai, Y. Dong, and H. Dong, A trust-based
probabilistic recommendation model for social networks, Journal of
Network and Computer Applications, vol. 55, pp. 5967, 2015.
[23] M. Han, M. Yan, Z. Cai, and Y. Li, An exploration of broader influence
maximization in timeliness networks with opportunistic selection, Jour-
nal of Network and Computer Applications, vol. 63, pp. 3949, 2016.
[24] D. Yang, G. Xue, X. Fang, and J. Tang, Crowdsourcing to smartphones:
Incentive mechanism design for mobile phone sensing, in Proceedings
of the 18th annual international conference on Mobile computing and
networking, 2012, pp. 173184.
[25] M. H. Cheung, R. Southwell, F. Hou, and J. Huang, Distributed time-
sensitive task selection in mobile crowdsensing, in Proceedings of the
16th ACM International Symposium on Mobile Ad Hoc Networking and
Computing, 2015, pp. 157166.
[26] D. Monderer and L. S. Shapley, Potential games, Games and economic
behavior, vol. 14, no. 1, pp. 124143, 1996.
[27] C.-J. Ho and J. W. Vaughan, Online task assignment in crowdsourcing
markets. in AAAI, vol. 12, 2012, pp. 4551.
[28] C.-J. Ho, S. Jabbari, and J. W. Vaughan, Adaptive task assignment for
crowdsourced classification, in Proceedings of the 30th International
Conference on Machine Learning (ICML-13), 2013, pp. 534542.
[29] Z. Duan, M. Yan, Z. Cai, X. Wang, M. Han, and Y. Li, Truthful incen-
tive mechanisms for social cost minimization in mobile crowdsourcing
systems, Sensors, vol. 16, no. 4, p. 481, 2016.
[30] Y. Singer, Budget feasible mechanisms, in Foundations of Computer
Science (FOCS), 2010 51st Annual IEEE Symposium on, 2010, pp. 765
774.
[31] A recursive algorithm to find all paths between two giv-
en nodes in c++ and c#, http://www.technical-recipes.com/2011/
a-recursive-algorithm-to-find-all-paths-between-two-given-nodes/, ac-
cessed: 2017-05-30.
[32] T. S. Ferguson, Who solved the secretary problem? Statistical science,
pp. 282289, 1989.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2017.2751304, IEEE Access

10

Zhuojun Duan received the MS degree from Qilong Han received the PhD degree in Computer
Shaanxi Normal University in 2011. Ms. Duan Science from Harbin Institute University, Harbin,
is currently a PhD student in the Department of China, in 2006. Dr. Han is currently a professor and
Computer Science at Georgia State University. Her Deputy Dean in the College of Computer Science
research interest is wireless network and social net- and Technology, Harbin Engineering University. His
work. research interests include data security and privacy,
mobile computing, distributed and networked sys-
tems. He has more than 70 publications as edited
books and proceedings, invited book chapters, and
technical papers in refereed journals and confer-
ences. He is a senior member of CCF, and the Chair
of CCF YOCSEF Harbin. He has served as programme committee members
and co-chairs of a number of international conferences/workshops for areas
including web intelligence, e-commerce, data mining, intelligent systems, etc.

Ling Tian received the bachelor, master and Ph.D.

degrees from the school of computer science, U-
niversity of Electronic Science and Technology of
China in 2003, 2006 and 2010, respectively. She is
currently an Associate Professor in UESTC. She had
been a visiting scholar in Georgia State University
during 2013 in United States. Her research interests
include image and video coding, signal processing.

Mingyuan Yan received the BS degree in Computer

Science and Technology and the MS degree in In-
formation Security from Wuhan University, Wuhan,
China, in 2008 and 2010, respectively. In 2012,
Dr. Yan received another MS degree in Computer
Science from Georgia State University. Dr. Yan is
currently an assistant professor in Computer Science Guisheng Yin received the Ph.D. degree in auto-
at University of North Georgia. Her research inter- matic control from Harbin Engineering University,
ests include data management and protocol design where he is a Full Professor and Doctoral tutor, the
in wireless networks, influence maximization and in- Dean of College of Computer Science and Technolo-
formation dissemination in mobile social networks. gy and the Dean of School of Software Engineering.
She is also interested in other topics such as information security and big data He ever worked in Tokyo University before he joined
management. Dr. Yan is an IEEE member, and an IEEE COMSOC member. the current university. His research interests are trust-
worthy software, information security, Internetware
and so on. He has published more than 100 papers
in well known journals and conferences. In addition,
he has presided 4 national projects and 5 provincial
projects.

Zhipeng Cai received his PhD and M.S. degrees in

the Department of Computing Science at University
of Alberta, and B.S. degree from Beijing Institute
of Technology. Dr. Cai is currently an Assistant
Professor in the Department of Computer Science
at Georgia State University. Dr. Cais research areas
focus on Networking, Privacy and Big data. He has
published more than 50 journals papers, including
more than 20 IEEE/ACM Transactions papers, such
as IEEE Transactions on Knowledge and Data Engi-
neering, IEEE Transactions on Dependable and Se-
cure Computing, IEEE/ACM Transactions on Networking, IEEE Transactions
on Mobile Computing, etc. Dr. Cai is the recipient of an NSF CAREER
Award. He is an editor/guest editor for Algorithmica, Theoretical Computer
Science, Journal of Combinatorial Optimization, and IEEE/ACM Transactions
on Computational Biology and Bioinformatics. He is a senior member of the
IEEE.

2169-3536 (c) 2017 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.