Wireless Sensor Network, 2011, 3, 174-182
doi:10.4236/wsn.2011.35020 Published Online May 2011 (http://www.SciRP.org/journal/wsn)
Towards Effective Bus Lane Monitoring Using
Camera Sensors
Xu Li1, Xuegang Yu2*, Ke He3
1Department of Computer Science and Engineering, State University of New York, Buffalo, USA
2College of Computer Science and Technology, Jilin University, Jilin, China
3Institute of Information Photonics and Optical Communications, Beijing University of Posts and
Telecommunications, Beijing, China
Received February 11, 2009; revised March 23, 2009; accepted April 1, 2009
Abstract
City administrators need to guarantee bus priority in urban public transportation. Building large-scale dedi-
cated bus lanes is a cost-effective solution but it suffers from illegal utilization of dedicated bus lines by
other non-permitted vehicles. In general, two systems can be utilized for bus lane monitoring: road-side sys-
tem and bus mounted system. Although the former one has the advantage in terms of larger surveillance
coverage, the investment cost makes it less feasible because of scalability issue. In this paper, we focus on
bus mounted system to improve surveillance coverage without additional infrastructure cost. We introduce
DoubleChecking, a cooperative violator identification scheme that can accurately pick out those non-per-
mitted vehicles or violators. DoubleChecking is designed to improve the surveillance coverage of bus
mounted system by using communications/cooperation between mounted camera sensors and existing cam-
era sensors around intersections. Through theoretical analysis and simulation results, we show that Dou-
bleChecking yields good performance for violator identification.
Keywords: Cooperative Violator Identification, Bus Lane Enforcement System, Sensors, ITS
1. Introduction
The rapid growth of modern society leads to an increas-
ing demand for advanced public transit system. Bus
Rapid Transit (BRT) is proposed to support bus-priority
transportation, but it needs considerable infrastructure
costs [1]. Another low-cost approach is to build dedi-
cated bus lanes in urban area. However, the illegal utili-
zations by non-permitted vehicles (violators) degrade the
effectiveness of dedicated bus lanes [2]. Fortunately,
intelligent transportation systems (ITS) aim to use vari-
ous modern technologies to improve road safety and ur-
ban traffic management [3-6].
In particular, bus lane enforcement systems have been
introduced to identify the violators. A typical approach
for violator identification is to use cameras and a key
component in this system is the Number Plate Reader
(NPR), by which the registration number of vehicle can
be analyzed based on image processing. In general, there
are two approaches to deploy the camera sensors:
road-side mounted system and bus mounted system. In
road-side mounted system, cameras are equipped in
lampposts along the road, which could capture violators
passing through bus lanes. Bus mounted system utilizes
the cameras mounted on buses to identify and record
these violators. The disadvantage of road-side mounted
system is high infrastructure cost while the drawback of
bus mounted system is the limited surveillance coverage.
In this paper, we are concerned with bus mounted sys-
tem and interested in whether the surveillance coverage
of bus mounted system can be increased without addi-
tional infrastructure cost. We propose DoubleChecking,
a cooperative scheme for violator identification in bus
lane enforcement system. Our solution utilizes the exist-
ing cameras around intersections and bus mounted cam-
eras to enhance the probability of identifying the real
violators. Specifically, a cooperative violator identifica-
tion scheme with probabilistic guarantee is adopted
based on wireless communications between cameras.
Thus, we can identify not only the violator immediately
in front of the bus (The number plate can be read directly,
e.g., vehicle A in Figure 1), but also the violator not

X. LI ET AL.
175
Figure 1. Application scenario of bus lane enforcement.
close to the bus (The buses cannot read the violator’s
number by bus mounted cameras directly because of
sight blocking, e.g., vehicle B in Figure 1). Dou-
bleChecking essentially extends the surveillance cover-
age of bus mounted system without additional infra-
structure cost. Overall, we mainly focus on novel archi-
tecture of bus lane enforcement system and the detailed
image-processing issue is beyond the scope of this paper.
The remainder of the paper is organized as follows.
Section 2 surveys the related work. Section 3 describes
the assumptions, model and problem statement. In Sec-
tion 4, we propose a novel DoubleChecking scheme for
cooperative violator identification. Section 5 is the per-
formance evaluation part, followed by the conclusion in
Section 6.
2. Related Work
Recently, a new infrastructure named Vehicular Sensor
Network (VSN) has attracted tremendous interests from
both government and academic. VSN is a network of
mobile sensors equipped on vehicles, such as taxis and
buses [3,7-9],which can be used for urban sensing, such
as traffic monitoring[10,11]. VSNs facilitate collection
of surveillance data over a wider area than the fixed in-
frastructure [6,12]. Meanwhile, unlike traditional wire-
less sensor networks, vehicular sensors are typically not
affected by strict energy constraints and vehicles can be
equipped with powerful processing units and wireless
transmitters. Actually, the work presented in this paper
still falls into vehicular sensor networks do- main and the
bus mounted cameras can be regarded as sensors.
Four bus lane enforcement projects have been carried
out in United Kingdom: one in Birmingham and the
other three in London [2,13,14]. The Birmingham Bus
Lane Enforcement System depends on video camera and
image processing equipment mounted on buses or along
the roadside. The camera digitally takes photos for the
vehicles in the bus lane twenty meters ahead of it and
transmits the numbers of those vehicles to an E-display
screen at the end of the bus lane. With this system, the
bus lane offences decreased by 60% and average bus
journey times decreased by 32% in Birmingham [14,15].
The Bus Lane Violation Detection & Deterrent (BLVDD)
in Heathrow is very similar to the roadside mounted sys-
tem in Birmingham, which is installed on a highway to
the airport [16]. The other two systems in London area
rely on SVHS video cassette recorder rather than digital
images to generate redundant information, with which
more accurate violator identification can be expected
[13]. Work in [17,18] studied the bus lanes enforcement
with intermittent priority, in which the author proposed a
cost effective approach to increase bus transit system
speed and reliability without creating excessive delays to
private vehicles.
3. Scenario, Model and Problem Statement
3.1. Scenario and Road Section Model
Actually, lots of buses are already equipped with various
electronic devices by bus companies, such as GPS, stor-
age device, wireless transmitter, etc. Besides that, we
assume that two forward facing cameras (color and
monochrome) and image processing unit are available on
each bus. With a GPS system, a bus can determine
whether or not it is in a bus lane and therefore whether or
not to open the two bus mounted cameras [2]. At the
same time, the existing static surveillance system around
the intersections (including similar devices mentioned
above) can be utilized to assist the bus mounted system
for violator identification without additional infrastruc-
ture cost. Especially, a color camera provides a wide
context view in front of the bus and a monochrome cam-

176 X. LI ET AL.
era shows a close up view of the vehicle’s number plate.
An infra-red illuminator enables operation in poor light-
ing conditions. With an image processing unit, the regis-
tration number of vehicle can be analyzed. Due to sight
blocking of bus mounted monochrome camera, it just
could read the number of violator immediately in front of
the bus. For the violators not close to the bus, only lim-
ited attribute information can be obtained by the wide
context view of the bus mounted color camera, such as
vehicle type, color, etc. However, the existing static
cameras around intersections are capable of reading all
the information of vehicles because of good viewpoint
(high position, good orientation, as shown in Figure 1).
In addition, with wireless communications, information
can be exchanged between intersections or between
buses and intersections to support cooperative violator
identification.
In this work, we use a typical road section: a road sec-
tion has three lanes and one of them is the dedicated bus
lane (as illustrated in Figure 1). Meanwhile, we assume
heavy traffic on the two public lanes and light traffic on
the bus lane (It is not worth and necessary for public ve-
hicle to be a violator if all lanes have similar traffic).
3.2. Problem Statement
In this work, we discuss the violator identification prob-
lem based on one-one model, i.e., a standard case unit
includes one road section and one bus running on this
road section. Accordingly, in each run of simulation, we
only focus on one standard case unit. Specifically, we
consider the following problem: it is known that we have
a bus mounted system and an existing intersection sys-
tem as introduced in Section 3.1, for a standard case unit,
how to sort out a suspicious vehicle (to be the violator)
set S by cooperative violator identification?
First, we introduce some definitions as follows.
Definition 1 (Maximum photographic distance): the
definition is mainly used for bus mounted cameras,
which have a limited field of vision. That is, they only
can generate high quality images for limited distance in
front of the bus. This parameter is named as maximum
photographic distance, denoted as mpd. How to regulate
the orientation of camera for capturing image and how to
deal with the image processing are beyond the scope of
this paper, related works can be found in [19-21].
Definition 2 (Average accuracy ratio, AAR): for a
standard case unit, we get a suspicious vehicle set S by
cooperative violator identification. The accuracy ratio
(AR) is defined as the proportion of real violator in the
set S. Thus, for W standard case units, the average ac-
curacy ratio is:
1
1W
j
j
j
S
AAR WS


 (1)
where
j
S and
j
S

are the sizes of suspicious vehicle
set S and the real violator vehicle set 
S
in the jth stan-
dard case unit, respectively. The metric indicates the
effectiveness of bus lane enforcement system. In addition,
we define another vehicle set  
S
=SS, which in-
cludes the vehicles which are in the suspicious vehicle
set but are not the real violators.
4. Cooperative Violator Identification for
Bus Lane Enforcement
4.1. Methodology
On one hand, some attribute information of vehicles
which are in mpd distance of a bus, can be identified
based on image processing. Typically, attribute informa-
tion includes vehicle type, color, brand, length, taxi or
private vehicle, etc. To be general, in the paper, we as-
sume that vehicle type and color information can be ob-
tained. Meanwhile, the number of vehicle immediately in
front of the bus can be obtained by monochrome camera.
In addition, the bus can estimate the relative locations of
violators in the whole traffic flow by GPS coordinates.
All related information can be transmitted from bus to
the forward intersection’s processing unit by wireless
communication, such as DSRC.
On the other hand, a road section is connected with
two intersections. With utilization of cameras at two in-
tersections, for a given unidirectional traffic flow, we can
construct two vehicle sequences based on their time-
stamps of entering/exiting the road section, respectively
(Accordingly, two indexes will be assigned to each vehi-
cle in both entering/exiting sequences and large index
corresponds to large timestamp). Basically, the two in-
dexes of a vehicle should not have considerable change
in two sequences due to the heavy traffic on the public
lanes, which is similar with FIFO. However, for an ex-
ceptional vehicle which is with large index in entering
sequence while small index in exiting sequence (i.e.,
compared with other vehicles, it took less time to travel
through the road section), it has higher probability to
have illegally utilized the bus lane as a violator.
Finally, with cooperation between intersections and
buses, a fraction of traffic flow can be selected based on
the location information of violators provided by bus.
Then, the exceptional vehicles in this targeted traffic
flow will be examined by utilizing more information sent
from the bus, such as vehicle type and color, etc.

X. LI ET AL.
177
4.2. DoubleChecking Scheme
We first give the definition of exceptional vehicle.
Definition 3 (Exceptional vehicle): for a given thresh-
old
 , the vehicle v will be regarded as an exceptional
vehicle if and only if:
v.eni – v.exi ≥ 
where v.eni and v.exi are the two indexes of v in the en-
tering and exiting sequences, respectively. Actually, here
we are only interested in the exceptional vehicles which
have less travelling time costs than others. The threshold
 is an empirical parameter based on traffic flow model,
which indicates how sensitive for the DoubleChecking
scheme to define exceptional vehicles.
Based on the discussion in Section 4.1, we propose
DoubleChecking, a novel cooperative violator identifica-
tion scheme for Bus Lane Enforcement, as shown in
Figure 2. In addition, we admit there are still exceptional
vehicles or violators which cannot be monitored by buses
because of limited mpd of bus mounted cameras. Actu-
ally, with increase of bus density, we may enable the
communication and cooperation between buses, which
will further improve the surveillance coverage of bus
lane enforcement system. Meanwhile, the model used in
this paper can also be extended so that the bus may
Scheme D OUBLE C HECKING :
// For a given bus b and road section with length l in standard case
unit
For a bus b, the bus mounted cameras take photographs for
the violators (denoted as v
i
) in its maximum photographic
distance. For each violator v
i
, with utilization of GPS and speed
estimator, the bus can calculate the time instant T
i
, at which v
i
will arrive at the forward intersection (corresponding to
parameter T
begin
at the intersection). Other information can be
also analyzed by image processing, such as vehicle type, color.
Especially, the number plate can be read if the violator is
immediate in front of the bus (denoted as v'). Then, All the
related information, e.g. (v
i
, type, color), will be transmitted to
the forward intersection’s processing unit by wireless
communication.
For the forward intersection of bus b, currently there are two
time-related parameters: T
begin
and T
end
. If the intersection
receives a new time instant T
j
from bus b and T
j
is earlier than
T
begin
, then T
begin
will be updated by T
j
. T
end
is the time instant, at
which v' will arrive at the forward intersection.
The traffic flow between [T
begin
, T
end
] will be examined for
violator identification. The exceptional vehicles in this traffic
flow will be sorted out based on Definition 3, which constitute a
vehicle set, denoted as S
0
. (First Checking Process)
For each vehicle in set S
0
, it will be added into final
suspicious vehicle set S if it has same (type, color) combination
provided by bus b. (Second Checking Process)
Figure 2. The DoubleChecking scheme.
monitor the violators behind it. Then, according scheme
can be designed based on DoubleChecking. We state this
part as our future work.
5. Performance Evaluation
5.1. Theoretical Analysis
In this section, we carried out a theoretical analysis about
expectation of accuracy ratio. For a standard case unit,
we have the following parameters:
N: number of vehicles in the whole traffic flow;
α: the proportion of exceptional vehicle in the traffic
flow (not including the real violator), this parameter is
related to the parameter;
n: number of real violators in suspicious vehicle set S;
tp: number of vehicle types;
clr: number of vehicle colors;
β: the proportion of traffic flow used for violator iden-
tification by DoubleChecking. Actually, this parameter is
related to the parameters Tbegin and Tend (Step 3 in Figure
2).
Accordingly, the number of exceptional vehicles in the
targeted traffic flow is N  α  β, which is denoted as m.
Now, we tend to calculate

E
S
and

EAR

, first
we have:








EEEEnE

 
 
S
SSS SS




 
112 2EP P
PiiPm
 
m

 
 


SS S
SS
Here, i



S means there are i out of m exceptional
vehicles in the suspicious vehicle set S. P(i


S) is the
probability that the incident i
 S happens. For given
numbers of color clr and vehicle type tp, there are totally
tp  clr combinations (In our work, we simply assume
that each vehicle randomly chooses its type and color).
Then, we have:







 
111
jjj
iii
PiPAPBA
PAPB A
PA PBA


 



S

where incident Aj means i exceptional vehicles occupy j
color  type combinations and incident Bj means that for
each of j color  type combinations in Aj, there exists real
violator in the suspicious vehicle set S, which has the
same combinations.
Theorem 1: the probability that incident Aj happens is:



11i jij
clr tpclr tp
ji
CjC j
PA
clr tp


 
 (2)

178 X. LI ET AL.
Here, the number of j-combinations (each of size j)
from a set with clr  tp elements (size clr  tp) is defined
as:


!
!!
j
clr tp
clr tp
Cjclr tpj



Proof: we use mathematical induction to proof our
statement.
Basis: the Equation (2) holds for j = 1.
 
11 0
1
10
ii
clr tpclr tpclr tp
ii
CC C
PA
clr tpclr tp
 
 


Inductive step: assume Equation (2) holds for j  q:



11, 1, 2,,
i
jij
clr tpclr tp
ji
CjC j
PAj q
clr tp


 














1
1
1
11
1
11
1
11
11
i
qq
clr tp
qx
i
x
ii
qxix
q
clr tpclr tpclr tp
i
x
ii
qqiq
clr tpclr tpclrtp
ii
Cq
PA PA
clr tp
Cq CxCx
clr tpclr tp
Cq CqCq
clr tpclrtpclrtp











 



 






i
i






 

12
211
1
1
1
12
2
1
ii
qq
clr tpclr tp
ii
i
clr tpclr tpclr tp
ii
i
qqi
clr tpclr tp
ii
xi
clr tpclr t
i
xq
CqCq
clrtpclrtp
CCC
clr tpclr tpclr tp
Cq Cq
clr tpclr tp
Cx C
clr tp








 







  

 


 










i


 
1
1
11
xi
p
i
xq
i
qqi
clr tpclr tp
ii
x
clr tp
Cq Cq
clrtpclr tp





 



Since both the basis and the inductive step have been
proved, it has now been proved by mathematical
induction that Equation (2) holds for all j. 
Theorem 2: the conditional probability of incident Bj
happens given Aj is:


1
1
11
i
jxx
jj j
x
clr tpx
PB ACclrtp


 
 



Proof: we knew that incident Aj means i exceptional
vehicles occupy j color  type combinations, now we
define incident Cl means there is no real violator which
occupies lth (l = 1, 2,

, j) in j color  type combinations
occupied by i exceptional vehicles. Thus,
Incident means there is at least one of j
1
j
l
l
C


color  type combinations, which is not occupied by a
real violator. Then we have,

1
1
j
j
jl
i
PB APC


 


 



11 1
1
1
12
1
jj
ll
llmj l
lmn
lmn j
j
j
P CPCPCC
PCCC
PCC C
 





 lm


 



Actually, it is easy to calculate:





 
12
, ,
llmlmn
PC PCCPCCCPCC...C j
For example,

1
1i clr tp
PC clr tp
 




12
2i clr tp
PCC clr tp
 




12
i
j
clr tpj
PCC Cclrtp
 
 




1
1 1
111
i
jjxx
jj ij
x i
clr tpx
PB APCCclrtp

 
 

 
 

 
We proved the theorem 2.
Then we have:
 





1
11
1
11 1
1
1
11
m
i
mi
jjj
ij
i
jij
clr tpclr tp
i
mi
i
j
ij xx
j
x
EPii
PAPB Ai
CjC j
clr tp
i
clr tpx
Cclr tp




 

 







 














  










SS

X. LI ET AL.
179



 



1
11 1
1
1
11
i
jij
clrtpclrtp
i
mi
i
j
ij xx
j
x
En
EAR EnE
n
CjC j
clr tp
ni
clr tpx
Cclr tp


 


 



 










 

  

 





S
SS
(3)
We use a practical case unit as an example and the
empirical parameter setting is as follows: there are totally
200 (N) vehicles in this case, in which 5% (α) are excep-
tional vehicles. Specially, 10% (β) of whole traffic flow
is used for violator identification and the bus reports
three (n) vehicles as violators during travelling the road
section in this case. The cameras can differentiate five
(tp) vehicle types and five (clr) colors. Then, we cal-
culated Equation (3) with values mentioned above and
E(AR) is about 97%, which shows that high accuracy
ratio can be expected by using DoubleChecking scheme.
5.2. Simulation Verification
We developed a traffic simulator using JAVA program-
ming language. In each run of simulator, we focus on
one standard case unit including one road section and a
bus travelling on it, as shown in Figure 1. We generated
the crowd traffic flow on the public lanes while the light
traffic on the bus lane. Especially, the driving behaviors
of bus and violators will be randomly chosen, such as
speed, route, etc. Table 1 lists the default parameters
used for the experiments in our simulation. The default
values of experimental parameters are selected based on
field experience. In addition, each data point in the fol-
lowing figures is averaged over 50 runs.
From Figure 3 to Figure 6, we plot the performance
curves of DoubleChecking with different parameter set-
tings. Overall, we can see that the DoubleChecking
scheme has high average accuracy ratio (AAR) for vio-
lator identification. To be more precise, in the most of
testing cases, nearly 90% vehicles in the suspicious vehi-
cle set S are the real violators, which demonstrates the
effectiveness of DoubleChecking.
Figures 3 and 4 show the AAR curve as offered road
section length increases from 50 m to 500 m and maxi-
Table 1. The default values of experiment par amete r s.
Length of road section (l) 300 m
Max. Photographic dist.(mpd) 30 m
The proportion of exceptional vehicle(α) 15%
No. of vehicle types (tp) 5
No. of vehicle colors (clr) 5
Figure 3. AAR with different lengths of road section.
Figure 4. AAR with different mpd.
Figure 5. AAR with different type  color combinations.
mum photographic distance increases from 15 m to 45 m,
respectively. As shown in Figures 3 and 4, with increase
of road length and maximum photographic distance, we
can see a slow decrease in AAR. This phenomenon can
be explained with the fact that with larger road length or

180 X. LI ET AL.
Figure 6. AAR with different proportions of exceptional
vehicles.
maximum photographic distance, the bus will catch more
violators during traveling, which leads to more type 
color combinations involved. Finally, there is a higher
probability for an exceptional vehicle to be included in
the suspicious vehicle set S. Conversely, Figure 5 shows
the AAR curve as offered type  color increases from 3 
3 to 7  7. It is easy to understand that with increase of
type and color numbers, there is an increase in AAR be-
cause large type  color combinations leads to a lower
probability for an exceptional vehicle to have same (type,
color) with any of the violators, due to the theoretical
analysis in Section 5.1. In Figure 6, we can see that Dou-
bleChecking has a good performance with different val-
ues of parameter α. Specially, even if 20% vehicles in the
whole traffic flow are exceptional vehicles (α = 20%),
high average accuracy ratio can still be expected, which
shows the stability and robustness of our DoubleCheck-
ing scheme.
5.3. Further Improvement
As shown above, our DoubleChecking scheme already
has high average accuracy ratio for violator identification.
After that, we still can utilize a decision system as used
in bus lane enforcement systems of UK [15,16], to con-
firm the violators with more accuracy. In this system, we
add the vehicles to a blacklist, which have been defi-
nitely identified as violators by DoubleChecking. It is
easy to explain that a violator may often use the bus lane
at any site. Thus, information sharing will be beneficial
for global bus lane enforcement. If a vehicle in suspi-
cious vehicle set S cannot be definitely identified as a
violator by DoubleChecking but it matches a record in
the blacklist, it will be regarded as a violator. Finally, the
system will generate an information packet to be admis-
sible as evidence in the courts for each violator, includ-
ing the related images, time stamp, site description, etc.
5.4. More Discussion about Bus Lane
Enforcement and Practical Issues
In this section, we tend to discuss some practical issues
about bus lane enforcement in city urban area. Actually,
the goal of dedicated bus lane is to improve the effi-
ciency of city transportation system, especially for public
transit. By deploying large-scale dedicated bus lanes in
road network, we hope buses can travel with a higher
speed. We point out that, however, there are also other
factors which may decrease the mean speed of buses.
Here, the most disadvantaged factor is traffic lights at
intersections in city urban area. To validate the influence
of traffic light, we carried out a real field testing, which
is trip-based approach.
We design a route which crosses the urban area of
Shanghai and take a taxi to finish our trip. The testing is
carried out on May 29, 2007, we start from at 10:30 and
arrive at end point at 13:09 with the whole trip of 56 km
(Figure 7). The mean speed of whole trip is about 21.1
km/h. However, we calculate the sum of time cost due to
traffic light delays at intersections, which almost come to
82 minutes whereas total time cost is 159 minutes. From
the test, we see that nearly 51.2% of total time cost is
waiting for red light, and the mean speed will be 43.8
km/h if there is no traffic delay during the whole trip.
This fact demonstrated that traffic light has a consider-
able influence on travelling speed of vehicles in city ur-
ban area. Accordingly, it will partially impair the advan-
tages of dedicated bus lane system.
In addition, we admitted that the numerical investiga-
tions presented in this work are based on somewhat ide-
alistic conditions and how the system will operate in a
realistic setting still needs more effort. For instance, there
is the case that some violator may turn right or left
Figure 7. The influence of traffic light on mean travelling
speed in urban area.

X. LI ET AL.
181
and choose different routes so that they cannot be cap-
tured by the cameras at the fixed positions. The probabil-
ity of this to happen could be capture in the mathematical
analysis. We state it as a part of our future work.
Several issues remain to be addressed further. Our fu-
ture work includes building a prototype system in
Shanghai and testing our DoubleChecking scheme on
this prototype. We hope the implementation experience
helps us further to understand the efficiency of Dou-
bleChecking. Second, DoubleChecking is a baseline
scheme which can serve as guideline when deploying
such an application in city urban area, how to design a
more sophisticated violator identification scheme is also
our future work (As discussed in Section 4.2, the model
can be extended that the violators behind the bus can also
be monitored. Meanwhile, we can enable the commu-
nication and cooperation between buses, which will fur-
ther improve the surveillance coverage of bus lane en-
forcement system). These works are currently in progress
in our lab.
6. Conclusions
We present a new scheme Doublechecking for bus lane
enforcement system, which is designed to identify the
violators in a cooperative manner. Compared with pre-
vious works, we tend to improve surveillance coverage
of bus mounted system without additional infrastructure
cost. To be more precise, we aim to identify not only the
violator immediately in front of the bus, bus also the
violators not close to the bus (The bus cannot read the
violator’s number directly because of sight blocking).
With DoubleChecking scheme, the violators can be
sorted out with high accuracy from traffic flow by the
cooperation between bus mounted cameras and the ex-
isting cameras around intersections. From both theoreti-
cal performance analysis and simulation results, Dou-
bleChecking shows a good performance for violator
identification, which demonstrates the feasibility of our
scheme.
7. Acknowledgements
This research was partially support by National Devel-
opment and Reform Commission under Grant No.
CNGI-09-01-11, and Jilin University under Grant No.
200903194.
8. References
[1] A. R. Girard, “Hybrid Supervisory Control for Real-Time
Embedded Bus Rapid Transit Applications,” IEEE
Transactions on Vehicular Technology, Vol. 54, No. 5,
2005, pp. 1684-1696.
doi:10.1109/TVT.2005.853466
[2] T. Ellis, “Deterring Bus Lane Bandits,” Traffic Technol-
ogy International Annual Review, 1998, pp. 192-194.
[3] U. Lee, E. Magistretti, M. Gerla, P. Bellavista and A.
Corradi, “Dissemination and Harvesting of Urban Data
Using Vehicular Sensor Platforms,” IEEE Transactions
on Vehicular Technology, Vol. 58, No. 2, 2009, pp.
882-901. doi:10.1109/TVT.2008.928899
[4] M. Sede, X. Li, D. Li and M.-Y. Wu, M. L. Li and W.
Shu, “Routing in Large-Scale Buses Ad Hoc Networks,”
IEEE Wireless Communications and Networking Con-
ference, Las Vegas, March 31-April 3, 2008, pp.
2711-2716. doi:10.1109/WCNC.2008.475
[5] Y. Yang and R. Bagrodia, “Evaluation of VANET-based
Advanced Intelligent Transportation Systems,” Proceed-
ings of the 6th ACM International Workshop on Vehicular
Internet Working, New York, 2009, pp. 3-12.
doi:10.1145/1614269.1614273
[6] H. Z. Zhu, M. L. Li et al., “SEER: Metropolitan-Scale
Traffic Perception Based on Lossy Sensory Data,” IEEE
INFOCOM 2009, Rio de Janeiro, 19-25 April 2009, pp.
217-225. doi:10.1109/INFCOM.2009.5061924
[7] B. Hull and V. Bychkovsky et al., “CarTel: A Distributed
Mobile Sensor Computing System,” Proceedings of the
4th International Conference on Embedded Networked
Sensor Systems, Boulder, October 31-November 3, 2006.
doi:10.1145/1182807.1182821
[8] J. Eriksson, H. Balakrishnan and S. Madden, “Cabernet:
Vehicular Content Delivery Using WiFi,” Proceedings of
the 14th ACM International Conference on Mobile Com-
puting and Networking, San Francisco, 14-19 September
2008, pp. 199-210. doi:10.1145/1409944.1409968
[9] A. Thiagarajan, L. Ravindranath et al., “VTrack: Accu-
rate, Energy-Aware Road Traffic Delay Estimation Using
Mobile Phones,” Proceedings of the 7th ACM Conference
on Embedded Networked Sensor Systems, Berkeley, No-
vember 2009. doi:10.1145/1644038.1644048
[10] X. Li, W. Shu et al., “Performance Evaluation of Vehi-
cle-Based Mobile Sensor Networks for Traffic Monitor-
ing,” IEEE Transactions on Vehicular Technology, Vol.
58, No. 4, 2009, pp. 1647-1653.
doi:10.1109/TVT.2008.2005775
[11] Xu Li et al., “Traffic Data Processing in Vehicular Sensor
Networks,” Proceedings of 17th International Confer-
ence on Computer Communications and Networks, St.
Thomas, 3-7 August 2008, pp. 1-5.
doi:10.1109/ICCCN.2008.ECP.42
[12] X. Li, H. Huang et al., “VStore: Towards Cooperative
Storage in Vehicular Sensor Networks for Mobile Sur-
veillance,” IEEE Wireless Communications and Net-
working Conference, Budapest, 5-8 April 2009, pp. 1-6.
doi:10.1109/WCNC.2009.4918021
[13] D. Turner and P. Monger, “The Bus Lane Enforcement
Cameras Project: The London Area Scheme,” Traffic En-
gineering & Control, Vol. 38, No. 10, 1997, pp. 529-539.
[14] S. Lewis, “The Bus Lane Enforcement Cameras Hand-
book (Provisional),” Home Office, St Albans, 1996.

X. LI ET AL.
182
[15] A. Wiggins, “Birmingham Bus Lane Enforcement Sys-
tem,” 9th International Conference on Road Transport
Information & Control, 21-23 April 1998, pp. 80-84.
doi:10.1049/cp:19980159
[16] G. Hill, “Bus Lane Violation Detection/Deterrent BLVDD,”
BAA Heathrow, 1998.
[17] M. D. Eichler, “Bus Lanes with Intermittent Priority:
Assessment and Design,” Masters of City Planning The-
sis, Department of City and Regional Planning, Univer-
sity of California, Berkeley, 2005.
[18] M. D. Eichler, “Bus lanes with Intermittent Priority:
Screening Formulae and an Evaluation,” Working Paper
UCB-ITS-VWP-2005-2, UC Berkeley Center for Future
Urban Transport, 2005.
[19] S. Greenhill and S. Venkatesh, “Distributed Query Proc-
essing for Mobile Surveillance,” Proceedings of the 15th
International Conference on Multimedia, Augsburg,
24-29 September 2007, pp. 413-422.
doi:10.1145/1291233.1291331
[20] B. Scheuermann, “A Fundamental Scalability Criterion
for Data Aggregation in VANETs,” Proceedings of the
15th Annual International Conference on Mobile Com-
puting and Networking, Beijing, 20-25 September 2009,
pp. 285-296. doi:10.1145/1614320.1614352
[21] S. Greenhill and S. Venkatesh, “Virtual Observers in a
Mobile Surveillance System,” Proceedings of the 14th
Annual ACM International Conference on Multimedia,
Santa Barbara, 23-27 October 2006, pp. 579-588.
doi:10.1145/1180639.1180759