• 沒有找到結果。

Surveillance on-the-road: Vehicular tracking and reporting by V2V communications

N/A
N/A
Protected

Academic year: 2021

Share "Surveillance on-the-road: Vehicular tracking and reporting by V2V communications"

Copied!
10
0
0

加載中.... (立即查看全文)

全文

(1)

Surveillance on-the-road: Vehicular tracking and reporting

by V2V communications

Lien-Wu Chen

a,⇑

, Yu-Chee Tseng

b

, Kun-Ze Syue

b a

Department of Information Engineering and Computer Science, Feng Chia University, Taichung 407, Taiwan

b

Department of Computer Science, National Chiao Tung University, Hsinchu 300, Taiwan

a r t i c l e

i n f o

Article history: Received 11 June 2013

Received in revised form 10 March 2014 Accepted 17 March 2014

Available online 18 April 2014 Keywords:

Dedicated Short Range Communications (DSRC)

License plate recognition (LPR) Vehicular network

Vehicular surveillance Video surveillance

a b s t r a c t

Vehicular networks have attracted a lot of attention recently. One potential application of vehicular networks is to use video cameras embedded in vehicles to support video surveil-lance, which we call ‘‘surveillance on-the-road’’. Traditional surveillance systems only rely on fixed stations on the roads to monitor road conditions. With vehicular cameras, deeper and richer road conditions may be tracked. In this paper, we study the related communi-cation issues to support such ‘‘on-the-road’’ surveillance scenarios. We use monitoring and tracking suspicious vehicles (such as stolen cars) on the road through license plate rec-ognition (LPR) as an example (our results should be applicable to other scenarios as well). We show how vehicles can work cooperatively through vehicle-to-vehicle (V2V) communi-cations to achieve this goal. With a tracking and a reporting modules, our solution does not rely on infrastructure networks. The tracking module allows handoff of a tracking job to neighboring vehicles as necessary and report of suspicious vehicles to nearby police cars. The reporting module can help guide message flows to avoid flooding the network. Simu-lation results verify the message efficiency of our approach. We also show how our frame-work can be applied to the developing WAVE/DSRC (Wireless Access in Vehicular Environments/Dedicated Short Range Communications) standards.

Ó 2014 Elsevier B.V. All rights reserved.

1. Introduction

Recently, lots of progress has been made in video sur-veillance due to the advances of video technologies and embedded computing and communications technologies. Many vehicular surveillance applications have been stud-ied, such as vehicle security[1], brake warning[2], and urban monitoring [3]. Traditional surveillance systems only rely on fixed stations on the roads to monitor road conditions. In this paper, we propose to use cameras embedded on vehicles to help the surveillance job, which

we call ‘‘surveillance on-the-road’’. Therefore, in addition to fixed video cameras, millions of cameras installed on vehicles may help the surveillance job. For example, most vehicles nowadays have rear cameras to assist backward driving. With vehicular cameras, deeper and richer road conditions may be tracked.

In this work, we study the communication issues to support such ‘‘on-the-road’’ surveillance scenarios through vehicle-to-vehicle (V2V) communications. We use tracking and reporting suspicious vehicles as an example. Most existing works [4–7] rely on roadside infrastructures to achieve this goal.

We propose an infrastructure-less framework for track-ing and monitortrack-ing suspicious vehicles by vehicles on the roads. Our solution consists of a tracking module and a reporting modules. The tracking module allows a vehicle http://dx.doi.org/10.1016/j.comnet.2014.03.031

1389-1286/Ó 2014 Elsevier B.V. All rights reserved.

⇑ Corresponding author. Tel.: +886 4 24517250x3759; fax: +886 4 24516101.

E-mail addresses:lwuchen@fcu.edu.tw(L.-W. Chen),yctseng@cs.nctu.

edu.tw(Y.-C. Tseng),kzsyue@cs.nctu.edu.tw(K.-Z. Syue).

Contents lists available atScienceDirect

Computer Networks

(2)

to keep tracking an identified suspicious vehicle and handoff its tracking job to neighboring vehicles as necessary. Also, on detecting an intersection, it allows a tracking vehicle to report the intersection passed by the suspicious vehicle to nearby police cars even if it has no digital map information at hand. Note that if vehicles have digital maps installed in their onboard units, an intersection might be detected by the current GPS location and digital maps. Otherwise, the proposed intersection detection scheme will be used to detect an intersection without relying on digital maps.

The reporting module helps find the nearest police car at low message cost without relying on costly roadside infrastructures, such as roadside units (RSUs) and wireless local area network access points (WLAN APs). Instead, it only relies on V2V communications. Note that if both the tracking vehicle and the nearest police car have 3G/3.5G radio interfaces, the discovery for suspicious vehicles might be directly reported to the nearest police car via 3G/3.5G communications. Otherwise, V2V communica-tions will be used to report the discovery as proposed in this work.

On the other hand, we propose to enhance LPR (licence plate recognition) capability to a vehicular camera, so a suspicious license number which is to be tracked can be recognized locally in a vehicle. This could greatly reduce the communication overhead. Also, there is no need of large amounts of human labors to watch these videos. Only in rare situations, a vehicular camera needs to keep a video clip for future use (such as a proof of evidence). Note that such video clips may be even clearer than those taken by fixed stations. Finally, our work should be applicable to the developing WAVE/DSRC (Wireless Access in Vehicular Environments/Dedicated Short Range Communications) stan-dards. We will comment how our work relates to WAVE/ DSRC.

Section 2 defines our suspicious vehicle tracking and reporting problem. Section3describes our framework to solve this problem. Simulation results are presented in Section4. Section5concludes the paper.

1.1. Related works

For tracking purposes, Ref.[4]presents an architecture for vehicle tracking systems using wireless sensor technol-ogies. RSUs are installed along roads to continuously keep tracking vehicles at regular intervals. These RSUs are con-nected to the underlying wired infrastructure to receive queries from the central server and reply back with the necessary information. Ref.[5]presents a WLAN-based real time system for vehicle localization. The proposed solution uses a neural network trained with a map of received power fingerprints from WLAN APs surrounding the vehi-cle. In these two works, however, a large number of RSUs and WLAN APs must be installed on the roadside to pro-vide target information and received signal strengths to vehicles, respectively. Ref. [7] proposes a smart parking scheme in VANETs, which includes tracking stolen vehi-cles. When a thief drives a stolen vehicle along a road, all pass-by RSUs can detect the parking beacon sent from the moving vehicle. According to the parking lot’s identifier in the beacon, the position of the stolen vehicle can be

reported to the parking lot. Again, this is also achieved by the deployment of RSUs along roads.

For reporting purposes, Ref. [6] proposes a searching strategy called ANTS to locate a desired vehicle close to the query user based on the lost ant searching for its nest. ANTS is employed in ShanghaiGrid[8]consisting of a large number of local nodes installed at crossroads, which is responsible for storing vehicle information and accepting queries. However, deploying such local nodes on each intersection requires a dramatic number of RFID readers and wireless APs and thus is costly. More importantly, it may not be practical to construct infrastructure in subur-ban and rural areas.

For message rebroadcast and contention resolution, most of VANET-specific schemes[9–14]do not consider the difference of vehicles located on intersections and road segments to employ different broadcast and backoff strat-egies for those vehicles. In addition, the memories of vehi-cles encountered the target car are not further investigated for cooperative searching on the road.

2. System model

Fig. 1shows our system model. We use suspicious vehi-cle tracking and reporting as an example (our results are not limited to such applications). We consider vehicles on the roads, which form vehicular surveillance networks via V2V communications. Each vehicle is equipped with a GPS receiver and some video cameras (one possibility is to install an embedded camera on each corner of the vehi-cle [15]). These video cameras can help recognize the license plates of the vehicles immediately in front/rear of it. With license plate recognition (LPR) techniques, trans-mitting license plate images is not always necessary. For identifying suspicious vehicles, the police department may publish a list of license plate numbers to be tracked in a particular region. Then vehicles in that region can try to recognize these plate numbers and keep those video clips when there is a match. We assume that each vehicle has an onboard communication unit, such as a Wi–Fi[16]

(3)

radio interface operating in the ad hoc mode under the same Basic Service Set Identifier (BSSID) or an IEEE 802.11p[17]radio interface operating in the WAVE mode with low connection setup overhead. In the WAVE mode, as long as vehicles operate in the same channel and use the same wildcard BSSID (with all 1 s as its address), two vehicles can immediately communicate with each other upon encountering on the road without having to join a BSS. Vehicles transmit periodical beacons to exchange IDs and positions with neighboring vehicles, and TTL (time to live) is indicated in broadcasting messages to limit their ranges.

For tracking and reporting suspicious vehicles, we for-mulate the problem as follow. The radio interface on each vehicle has a fixed transmission range R. By LPR, each vehi-cle i can identify whether the immediate front/behind vehicle

v

f=

v

bis a suspicious vehicle

v

s or not. Below, we

use

v

f as an example (our framework should be applicable

to

v

bas well). Vehicle i recognizes

v

f’s license plate

num-ber every tu and tnseconds ðtu<tnÞ in the urgent mode

and normal mode, respectively. Before

v

shas been

identi-fied, i recognizes the license plate number of its

v

f in the

normal mode. Once

v

f is identified as

v

s;i will immediately

report this discovery to nearby police cars

v

pand switch to

the urgent mode to continuously recognize the license plate number of its

v

fto keep tracking

v

s. In addition, since

the tracked

v

scannot change its direction on road segment,

the current position of

v

s will be reported to

v

p on each

intersection for reducing the number of reporting mes-sages and maintaining the up-to-date location of

v

s. On

the other hand, if

v

sis not in front of i due to changes of

lane or direction, the tracking job will be handoff from i to i’s neighboring vehicle which is immediate behind

v

s.

Our goal is to design efficient protocols for vehicles to cooperatively track the identified

v

s and to report each

intersection passed by

v

sto

v

pduring the tracking process.

Such reporting messages mrshould be guided to the

near-est

v

p and delivered through multi-hop forwarding.

Through these mr;

v

p can reconstruct the trajectory of

v

s

and take further actions a.s.a.p. Given such scenarios, we will consider the following four research issues.

1. Tracking handoff: How do we handoff the tracking job to neighboring vehicles as

v

schanges its lane or

direc-tion so that

v

scan be continuous tracked on the road?

2. Intersection detection: How do we detect an tion without digital maps at hand so that each intersec-tion passed by

v

scan be reported to

v

pfor maintaining

the up-to-date location of

v

s?

3. Rebroadcast decision: How do we report the location of

v

sto nearby

v

psuch that the number of rebroadcasts of

mrcan be minimized?

4. Message guiding: How do we guide the propagation of mr to the nearest

v

p without flooding the network

so that the message overhead for reporting can be reduced?

3. Tracking and reporting protocols

In this section, we propose an infrastructure-less frame-work for the vehicle tracking and reporting problem, which

consists of a tracking module and a reporting module, as shown in Fig. 2. First, to keep tracking an identified

v

s,

we propose a tracking handoff scheme and an intersection detection scheme in Section3.1. Second, to efficiently deli-ver reporting messages about

v

sto the nearest

v

p, we

pro-pose a rebroadcast decision, an intersection-guiding search, and a memory-based backoff schemes in Section3.2.

Table 1summarizes the notations to be used.

3.1. Tracking module

We consider that vehicles use their onboard cameras to conduct surveillance in a cooperative manner, which we call vehicular surveillance networks. Through its onboard camera, each vehicle i can take snapshots on its

v

f. i

retrieves this license plate number and compares it against a database of suspicious plate numbers provided by the police department (through the Internet or nearby police cars). When a suspicious vehicle

v

sis identified, i will

con-tinuously take snapshots on

v

s. Note that to avoid privacy

leakage, the suspicious plate numbers from the police department could be the values returned by a hash func-tion. Similarly, the recognized plate number of

v

f could

be calculated using the same hash function and compared without knowing the actual plate number. For cooperative tracking, we design a tracking handoff scheme to pass the tracking job to a vehicle immediately behind

v

s due to

changes of lane/direction. In addition, for maintaining the up-to-date location of

v

s, we design an intersection

detec-tion scheme to report the current posidetec-tion of

v

s to

v

p on

each intersection during the tracking process. Note that since the tracked

v

s cannot change its direction on road

segment, the position of

v

s is only reported on each

intersection for reducing the number of reporting

messages to

v

p.

3.1.1. Tracking handoff scheme

If vehicle i detects its

v

f as

v

sduring the normal mode, i

will switch to the urgent mode and continuously track

v

s.

Once i becomes unable to track

v

sduring the urgent mode,

i will broadcast a tracking handoff message mh to

neigh-boring vehicles. Vehicles receiving mh will switch to the

v

s

v

s

v

p

(4)

urgent mode and continuously check if it can detect the missed

v

s. The neighboring vehicle j which detects

v

swill

take over the tracking job from i without replying any acknowledgment message to i, and all other neighboring vehicles will go back to the normal mode after a threshold (tracking handoff timer). Similarly, once j becomes unable to track

v

s, it will repeat the above procedure. Note that if

there is no neighboring vehicle that can take over the tracking job, the current tracking job is terminated and a new tracking process will be started again when another vehicle k detects its

v

f as the missed

v

sduring the normal

mode. InFig. 3, for example, A is a tracking vehicle in the urgent mode and B is a suspicious vehicle tracked by A. Once A becomes unable to track B; A will broadcast mh to

C and D. C and D will switch from the normal mode to the urgent mode for detecting B. On one hand, since C has detected B, the tracking job is handoff from A to C. On the other hand, after the tracking handoff timer expires, D will switch back to the normal mode.

3.1.2. Intersection detection scheme

Since vehicles driving on the roads will transmit period-ical beacons to exchange IDs and positions with neighbor-ing vehicles, each vehicle i can maintain the up-to-date locations of its neighboring vehicles Nibased on received

beacons (we consider only moving vehicles, not including parked cars). According to the positions of Ni, we remove

nearby Ni within a certain distance d from i (depending

on the average width of road segments) and divide Niinto

four parts, which are located on i’s head sector sH, tail

sec-tor sT, right sector sR, and left sector sLwith angles hH;hT;hR,

and hL, respectively. We set hH¼ hT;hR¼ hL, and

hH¼ 180 hR, where hHand hT can be determined based

on the average width of road segments. InFig. 4(a), for example, the radius of the inner circle is equal to d and the radius of the outer circle is equal to the distance from vehicle A and the farthest NA.

To detect if i is located on intersection, we check whether there is any Ni in sH;sT;sR, or sLof i. If there is Ni

in either sRor sL;i is located on intersection. If there is no

Niin sRand sL;i is located on road segment. The basic

con-cept is that only when i is located on intersection, it might receive beacons from Niin its sRor sL. InFig. 4, for example,

vehicle B detects that it is located on intersection because there are NBin its sRand sLwhereas vehicle A; C; D, and E

detect that they are located on road segment since there is no neighboring vehicles in their sR or sL. Note that the

moving directions of neighboring vehicles in sRor sL can

be also used for intersection detection in addition to their positions.

3.2. Reporting module

Once

v

s is identified by vehicle i, the discovery should

be reported to the nearest

v

p. If both i and

v

p have 3 G/

3.5 G radio interfaces, the discovery might be directly reported to

v

p via 3 G/3.5 G communications. Otherwise,

V2V communications will be used to report the discovery to

v

pas to be discussed below. After the nearest

v

preceives

mrreported by i, it can take less time to arrive at the

posi-tion of

v

sand take further action. Although flooding is an

intuitive search scheme that can always find the nearest

v

p, it causes a huge amount of network traffic and thus

makes its scalability low. Message mr contains the

dis-cover’s ID,

v

s’s position and license number, and a

sequence number. To reduce the message overhead for reporting mr to nearby

v

p, we design a rebroadcast

deci-sion, an intersection-guiding search, and a memory-based backoff schemes to minimize the number of rebroadcasts.

3.2.1. Rebroadcast decision scheme

In this scheme, vehicles decide whether they rebroad-cast mraccording to their positions, mr’s reporting

direc-tion dr, and the corresponding sector sC for dr. In

Fig. 4(b), for example, sender A is located on sT of receiver

B so that dris from sTto sHand thus sCfor dris sH.

Accord-ingly, sC for drof C; D, and E are sR;sT, and sR(i.e., shaded

areas in Fig. 4(c)–(e)), respectively. The rebroadcast decision is made as follows.

Table 1

Summary of notations. Notation Definition

vs The identified suspicious vehicle, such as a stolen car

vf The immediate front vehicle

vp The police car for dealing withvs

mh The tracking handoff message sent to neighboring

vehicles

mr The reporting message sent to nearbyvp

dr The reporting direction of mr

dg The guiding direction of mr

Ni The neighboring vehicles of vehicle i

sH The sector with angle hHlocated on the head of a vehicle

sT The sector with angle hTlocated on the tail of a vehicle

sR The sector with angle hRlocated on the right of a vehicle

sL The sector with angle hLlocated on the left of a vehicle

sC The corresponding sector for dr

tu The LPR interval in the urgent mode

tn The LPR interval in the normal mode

ti The passing time from vehicle i metvpto now

s The small integer for the first backoff window q The number of backoff classes

T The valid duration of the memory forvp

Fig. 3. An example of handoff a tracking job to neighboring vehicles as necessary.

(5)

1. When vehicle j receives mr sent by vehicle i; j will

first detect if it is located on intersection. If so, j will rebroadcast mr immediately. Otherwise, it will

detect sCfor drand then check if there is any Njin sC.

2. If there is Njin sC;j will rebroadcast mr. Otherwise, j

will discarded mrto avoid unnecessary rebroadcasts

since there is no Njthat can help rebroadcast.

Similarly, after vehicle k receives mrfrom j; k will repeat

above procedures. InFig. 4, for example, vehicle B and D decide to rebroadcast mrbecause B is located on

intersec-tion and there is ND in sC of D (i.e., sT), respectively,

whereas vehicle C and E decide not to rebroadcast mrsince

there is no NCand NEin their sCfor dr(i.e., sR). Note that the

reception of mrwith the same sequence number serves as

an implicit message to prevent other vehicles received mr

competing again. Therefore, if vehicle B rebroadcasts mr

received from vehicle A; NATNB will not rebroadcast A’s

mr.

3.2.2. Intersection-guiding search scheme

Based on the rebroadcast decision scheme, we further design an intersection-guiding search scheme for guiding mr to the nearest

v

p. In our scheme, each vehicle i will

record the position as it meets

v

p (through the beacon

broadcast by

v

p). When i receives mr, it will specify the

guiding direction dgin mraccording to its memory for

v

p.

mr will be rebroadcast to the recorded position of

v

p in

(6)

i’s memory (if the memory is still valid). Thus, mr can be

guided on intersection by vehicles’ memory for

v

p, so the

flooding of mrcan be avoided and the number of

rebroad-casts of mrcan be reduced. Note that the valid duration of

the memory for

v

pis bounded by a predefined threshold T

(which is set to 30 s in our simulation). Thus, the up-to-date position of

v

p can be updated by the vehicle with

the freshest memory (based on the scheme proposed in Section3.2.3). Even if mrmight be initially guided by the

memory not fresh enough, the message can still be guided to

v

psince the transmission speed of packets is much

fas-ter than the moving speed of vehicles. We illustrate an example inFigs. 5and6as follows.

1. InFig. 5, vehicle A identifies its

v

f as

v

s and

broad-casts mr to the nearest

v

p. There are four

intersec-tions traveled by mr, where vehicle B is located on

the first intersection, vehicle C and D are located on the second intersection, vehicle E is located on the third intersection, and

v

p is located on the fourth

intersection.

2. On the first intersection, as shown inFig. 6(a), vehi-cle B detects that it is located on intersection and mr

is from sT (i.e., from A to B). In addition, B did not

meet

v

p before. So B will decide to rebroadcast mr

without specifying dg (i.e., denoted by ‘‘broadcast

to: sC’’ instead of ‘‘sH’’, ‘‘sT’’, ‘‘sR’’, or ‘‘sL’’). Thus, other

vehicles located on the first intersection will decide their own dgby themselves as receiving mrfrom B.

3. On the second intersection, as shown inFig. 6(b), vehicle C detects that it is located on intersection and mr is from sH (i.e., from B to C). In particular,

C met

v

p10 s ago. So C will guide mrto its sL, which

is closest to the recorded position of

v

p 10 s ago.

Thus, only NC in sL of C will help rebroadcast mr

sent by C. NCin sH;sT, and sRwill discard mr

imme-diately because dg in mr has been specified to sL

(i.e., denoted by ‘‘broadcast to: sL’’), where dg

spec-ified by C is the relative direction between the posi-tion of C and the past posiposi-tion of

v

pin the memory

of C.

4. On the third intersection, as shown inFig. 6(b), and (d), vehicle E detects that it is located on intersection and mr is from sH (i.e., from D to E). Based on the

memory for

v

p of C; mr will be rebroadcast to sR of

E. However, E met

v

p5 s ago, which is more

up-to-date than C. So mr is guided to sLof E according to

the memory for

v

p of E. Therefore,

v

p can receive

mrin the next rebroadcast.

3.2.3. Memory-based backoff scheme

In the IEEE 802.11p standard, the Enhanced Distributed Channel Access (EDCA) originally provided by IEEE 802.11e is employed for prioritizing channel access [18]. It is accomplished by using different channel access parame-ters for each packet priority and there are four access cat-egories defined for background (AC_BK), best effort (AC_BE), video (AC_VI), and voice (AC_VO) traffic. A backoff scheme is adopted in EDCA, which consists of Arbitration Interframe Space Number (AIFSN) and a random backoff timer. AIFSN is a fixed waiting time in unit of slot, whereas the backoff timer is a random waiting time selected from a Contention Window (CW). The CW size is initially set to CWminand doubled until reaching CWmaxafter each

trans-mission collision. The default EDCA parameter set of IEEE 802.11p is shown inTable 2. In our scheme, mris assigned

to AC_VO with the smallest AIFSN and a memory-based backoff timer.

To reduce the number of rebroadcasts of mr, we design

a memory-based backoff scheme. It facilitates receivers

(7)

met

v

pmore recently to rebroadcast at earlier time. When

vehicle i receives mr;i first decides to rebroadcast mr or

not. If i decides to rebroadcast mr, a backoff timer will be

assigned to i based on the passing time ti from i met

v

p

to now. If i never met

v

p before, ti will be set to 1.

A smaller tiwill give a smaller backoff timer BTi, as defined

below: BTi¼ ½0; 2sþ1 1 0 < t i61qT ½2sþ1;2sþ2  1 1 qT < ti6q2T .. . ½2sþq1;2sþq 1 q1 q T < ti61 8 > > > > > > < > > > > > > : ; ð1Þ

where

s

is a small integer,

q

is the number of backoff clas-ses, and T is the valid duration of the memory for

v

p. Thus,

this gives receivers met

v

pmore recently higher priorities

to rebroadcast. Fig. 6. Example of the intersection-guiding search scheme.

Table 2

Default EDCA parameter set.

AC CWmin CWmax AIFSN

AC_BK aCWmin aCWmax 9

AC_BE (aCWmin+ 1)/2  1 aCWmin 6

AC_VI (aCWmin+ 1)/4  1 (aCWmin+ 1)/2  1 3

(8)

On the other hand, an implicit inhibition strategy[13]is adopted to eliminate redundant mr. Specifically, the

recep-tion of mr with the same sequence number serves as an

implicit message to prevent i from competing again. On receiving such a rebroadcasting, i will remove the message in its waiting queue. Furthermore, to improve reliability, a vehicle sending mrwill set its backoff timer to 2sþqand try

to overhear any rebroadcasting from any neighboring vehi-cle. If it cannot overhear any such rebroadcasting, it will

back off to 0 and rebroadcast mr again with a new

sequence number.

4. Performance evaluation

We simulate the proposed framework by QualNet 5.0

[19] with some modifications. Basic parameters used in our simulation are summarized inTable 3. A 5-km2urban area consisting of 25 1-km2 building blocks with 250, 500, 750, 1000, 1250, 1500, and 1750 vehicles is simulated, as shown inFig. 7. All vehicles are uniformly placed on each road segment and their moving directions are ran-domly selected from right-turn, left-turn, and go-straight on each intersection. Both the suspicious car and the police car are randomly chosen. We set tu¼ 1 s, tn¼ 10 s,

d¼ 10 m, hH¼ 60;

s

¼ 1;

q

¼ 3, and T ¼ 30 s. We compare

our scheme against the traditional flooding method and the intelligent broadcast method[9,13]that avoids redun-dant rebroadcasts by implicit acknowledgements. The main performance indices are total number of reporting messages, packet collision rate, and average reporting delay to the police car. Each simulation is repeated 100 times and then we take the average value.

Fig. 8 illustrates the total numbers of reporting mes-sages under different numbers of vehicles. We can observe that our scheme has the lowest number of reporting mes-sages. This is because our scheme could guide reporting messages on intersections so that they are only rebroad-cast to one of road segments based on the memory for the police car. On the contrary, the flooding scheme and the intelligent broadcast scheme will rebroadcast report-ing messages to all road segments on intersections so that their total numbers of reporting messages are much more than ours.

Fig. 9 shows the packet collision rate under different numbers of vehicles. It can be observed that our scheme Table 3

Simulation parameters.

Parameter Value

MAC protocol IEEE 802.11a

Radio model Two-ray ground

Routing protocol Broadcast forwarding Reporting message size 128 bytes

Beacon interval 1 s

Tracking handoff timer 10 s

Transmission range 300 m

Number of vehicles 250–1750 vehicles

Vehicle speed 40–60 km/h

Fig. 7. Manhattan topology used in the experiments. 0 10000 20000 30000 40000 50000 60000 250 500 750 1000 1250 1500 1750

Total number of reporting messages

Number of vehicles

Flooding Intelligent Broadcast Our Framework

Fig. 8. Comparison of total number of reporting messages.

0 0.2 0.4 0.6 0.8 1 1.2 1.4 250 500 750 1000 1250 1500 1750

Packet collision rate (%)

Number of vehicles Flooding Intelligent Broadcast Our Framework

(9)

still outperforms flooding and intelligent broadcast. The reason is similar to what is discussed earlier. In addition, our memory-based backoff scheme further reduces the packet collision rate because we prioritize reporting messages. For irregular road geometry, some intersections are randomly selected to assign the go-straight moving direction all the time to each vehicle passing them. The similar results are obtained that our scheme has the lowest number of reporting messages and the lowest packet collision rate under different numbers of vehicles.

Fig. 10shows the average reporting delay to the police car for various penetration rates of vehicles equipped with onboard communication units on the roads as the total number of vehicles is 1750. With 14.2% and 28.5% penetra-tion rates, the numbers of communicated vehicles are too less to report the discovery to the police car. As the pene-tration rate increases, our approach has the similar report-ing delay (and reportreport-ing success ratio) to floodreport-ing and intelligent broadcast while keeping the message cost low. In particular, when there is 100% of vehicles broadcasting periodical beacons every second, our approach slightly outperforms flooding and intelligent broadcast due to its less number of reporting messages. On the other hand,

Fig. 11 shows the handoff success rate of our tracking handoff scheme under different numbers of vehicles and tracking handoff timers. The number of vehicles and their handoff timers are varied from 250 to 1750 vehicles and 5 to 30 s, respectively. With more vehicles and larger tim-ers, the handoff success rate can increase from 13% to 97%.

Consequently, our proposed approach can achieve the lowest number of rebroadcasts and packet collision rate while keeping the reporting delay as low as flooding and intelligent broadcast. It is leading to more efficient use of wireless bandwidth in vehicular networks. In other words, adopting our approach can both avoid the overhead of reporting messages wasting bandwidth due to

unneces-sary rebroadcasts and prevent reporting messages

from transmission collisions caused by serious packet contention.

5. Conclusion

This paper proposes ‘‘Surveillance On-the-Road’’ that integrates surveillance technologies with WAVE/DSRC communications to support vehicle tracking and reporting applications. We use monitoring and tracking suspicious vehicles as an example and design an infrastructure-less framework consisting of a tracking and a reporting mod-ules. Vehicles can work cooperatively through V2V com-munications to achieve this goal. The tracking module can handoff a tracking job to neighboring vehicles as nec-essary and report suspicious vehicles to nearby police cars. The reporting module can help guide message flows to avoid flooding the network. In the paper, we mainly focus on how to design the vehicular tracking and reporting by V2V communications. The mathematical analysis for those proposed schemes will be further investigated in our future work. Simulation results show that our approach outperforms existing works in the message efficiency. Acknowledgments

This research is supported in part by NSC under Grant No. 102-2221-E-035-031-MY3. Y.-C. Tseng’s research is co-sponsored by MoE ATU Plan, NSC 101-2221-E-009-024-MY3, NSC 102-2218-E-009-002,

NSC102-2911-I-002-001, NTU103R7501, IVF-NSC joint research Grant

21280013 (102-2923-E-009-001-MY2), Academia Sinica AS-102-TP-A06, ITRI, hTC, Delta, and D-Link.

References

[1] L.-W. Chen, K.-Z. Syue, Y.-C. Tseng, VS3

: a vehicular surveillance and sensing system for security applications, in: IEEE International Conference on Mobile Ad-hoc and Sensor Systems (MASS), October 2009.

[2] P. Thammakaroon, P. Tangamchit, Adaptive brake warning system for automobiles, in: International Conference on ITS Telecommunications (ITST), October 2008, pp. 204–208.

[3]U. Lee, E. Magistretti, M. Gerla, P. Bellavista, A. Corradi,

Dissemination and harvesting of urban data using vehicular sensing platforms, IEEE Trans. Vehicular Technol. 58 (2) (2009)

882–901.

[4] K.G. Aravind, T. Chakravarty, M.G. Chandra, P. Balamuralidhar, On the architecture of vehicle tracking system using wireless sensor devices, in: International Conference on Ultra Modern Telecommunications & Workshops (ICUMT), October 2009. [5] M. Caceres, F. Sottile, M.A. Spirito, WLAN-based real time vehicle

locating system, in: IEEE Vehicular Technology Conference (VTC-Spring), April 2009, pp. 1–5.

[6]M. Li, H. Zhu, Y. Zhu, L.M. Ni, ANTS: efficient vehicle locating based

on ant search in ShanghaiGrid, IEEE Trans. Vehicular Technol. 58 (8)

(2009) 4088–4097.

[7] R. Lu, X. Lin, H. Zhu, X. Shen, SPARK: a new VANET-based smart parking scheme for large parking lots, in: IEEE Conference on 600 650 700 750 800 850 900 950 1000 14.2 28.5 42.8 57.1 71.4 85.7 100

Average reporting delay (ms)

Penetration rate (%)

Flooding Intelligent Broadcast Our Framework

Fig. 10. Comparison of average reporting delay to the police car.

10 20 30 40 50 60 70 80 90 100 250 500 750 1000 1250 1500 1750 5 10 15 20 25 30 10 20 30 40 50 60 70 80 90 100

Fig. 11. Handoff success rate under different numbers of vehicles and tracking handoff timers.

(10)

Computer Communications (INFOCOM), April 2009, pp. 1413– 1421.

[8]M. Li, M.-Y. Wu, Y. Li, ShanghaiGrid: an information service grid,

Concurr. Comput.: Pract. Exp. 18 (2006) 111–135.

[9]S. Biswas, R. Tatchikou, F. Dion, Vehicle-to-vehicle wireless

communication protocols for enhancing highway traffic safety,

IEEE Commun. Mag. 44 (1) (2006) 74–82.

[10] O. Tonguz, N. Wisitpongphan, F. Bai, P. Mudalige, V. Sadekar, Broadcasting in VANET, in: Proc. Mobile Networking for Vehicular Environments, May 2007, pp. 7–12.

[11] S. Khakbaz, M. Fathy, A reliable broadcast method for vehicular ad hoc networks considering fragmentation and intersection problems, in: International Conference on Next Generation Mobile Applications, Services and Technologies (NGMAST), September 2008, pp. 379–384.

[12] Y. Bi, L. Cai, X. Shen, H. Zhao, A cross layer broadcast protocol for multihop emergency message dissemination in inter-vehicle communication, in: IEEE International Conference on Communications (ICC), May 2010, pp. 1–5.

[13]L.-W. Chen, Y.-H. Peng, Y.-C. Tseng, An infrastructure-less framework

for preventing rear-end collisions by vehicular sensor networks, IEEE

Commun. Lett. 15 (3) (2011) 358–360.

[14]F.J. Ros, P.M. Ruiz, I. Stojmenovic, Acknowledgment-based broadcast

protocol for reliable and efficient data dissemination in vehicular ad

hoc networks, IEEE Trans. Mobile Comput. 11 (1) (2012) 33–46.

[15] LUXGEN’s Advanced Technologies. <http://www.luxgen-motor.com/

technology/technology-experience.asp>.

[16] WI-FI Alliance. <http://www.wi-fi.org>.

[17] IEEE std. 802.11p/D4.0, Draft Amendment for Wireless Access in Vehicular Environments (WAVE), March 2008.

[18] IEEE P1609.4/D6.0, Draft Standard for Wireless Access in Vehicular Environments (WAVE) – Multi-channel Operation, March 2010. [19] QualNet. <http://www.scalable-networks.com/products/qualnet/>.

Lien-Wu Chen received his B.S. and M.S. degrees both in Computer Science and Infor-mation Engineering from the Fu-Jen Catholic University and the National Central University in 1998 and 2000, respectively. He was a Research Assistant at the Institute of Infor-mation Science, Academia Sinica (Jan. 2001– Feb. 2009). He obtained his Ph.D. in Computer Science and Information Engineering from the National Chiao-Tung University in December of 2008. He joined the Department of Com-puter Science, National Chiao-Tung University as Postdoctoral Research Associate in Feb. 2009, and qualified as Assistant Research Fellow in Feb. 2010. He is currently an Assistant Professor since Feb. 2012 in the Department of Information Engineering and Computer Science, Feng-Chia University. His research interests include wireless communication and mobile computing, especially in Mobile Ad Hoc and Sensor Networks.

Dr. Chen has published 40 journal/conference papers and has 25 approved/filed patents. Among those patents, the technologies for file

transfer, application execution, and goods data searching methods based on Augmented Reality and Cloud Computing have been transferred to Acer Incorporated in Jan., June, and Nov. 2011, respectively. He was a guest researcher at the Industrial Technology Research Institute (Feb. 2009–Dec. 2009) and Chief Executive Officer at the RFID Resource Center, National Chiao-Tung University (Apr. 2009–Mar. 2011). Dr. Chen received the Outstanding Demo Award (1st author) at IEEE MASS 2009, the Best Paper Award (1st author) at WASN 2010, the Audience Award (2nd prize) at ESNC 2011, the Best Paper Award (1st author) at MC-2012, the Pro-totyping Award (Stars prize) at ESNC 2012, the Best Paper Award (1st author) at WASN 2013, the Best Demo Award (1st author) at MC-2013, and the Best Paper Award (1st author) at NCS 2013. He is an honorary member of Phi Tau Phi Scholastic Honor Society, a lifetime member of IICM, and a member of ACM and IEEE.

Yu-Chee Tseng got his Ph.D. in Computer and Information Science from the Ohio State University in January of 1994. He was/is Chairman (2005–2009), Chair Professor (2011–present), and Dean (2011–present), Department of Computer Science, National Chiao Tung University, Taiwan. Dr. Tseng is a Y.Z. Hsu Scientific Chair Professor. Dr. Tseng received Outstanding Research Award (National Science Council, 2001, 2003, and 2009), Best Paper Award (Int’l Conf. on Paral-lel Processing, 2003), Elite I.T. Award (2004), and Distinguished Alumnus Award (Ohio State University, 2005), and Y.Z. Hsu Scientific Paper Award (2009). His research interests include mobile computing, wireless communication, and sensor networks. Dr. Tseng is an IEEE Fellow. He serves/served on the editorial boards of IEEE Trans. on Vehicular Technology (2005–2009), IEEE Trans. on Mobile Computing (2006–2011), and IEEE Trans. on Parallel and Distributed Systems (2008–present).

Kun-Ze Syue received his B.S. and M.S. degrees both in Computer Science and Infor-mation Engineering from the Chinese Culture University and the National Chiao Tung Uni-versity in 2005 and 2010, respectively. Before obtaining the M.S. degree, he has ever worked as a software programmer and majored in user interface design (Nov. 2006–May 2008). He is currently a Research & Development Senior Engineer since Aug. 2010 at Alpha Networks Inc., Taiwan.

He received the Outstanding Demo Award at the 6th IEEE International Conference on Mobile Ad-hoc and Sensor Systems (MASS 2009). His research interests include vehicular ad hoc networks and wireless sensor networks.

數據

Fig. 1 shows our system model. We use suspicious vehi- vehi-cle tracking and reporting as an example (our results are not limited to such applications)
Table 1 summarizes the notations to be used.
Fig. 4 (b), for example, sender A is located on s T of receiver
Fig. 4. Rebroadcast decisions based on neighboring vehicle distributions.
+4

參考文獻

相關文件

Define instead the imaginary.. potential, magnetic field, lattice…) Dirac-BdG Hamiltonian:. with small, and matrix

Service learning: Connecting citizenship with the classroom. Assessing and reporting on habits

Service learning: Connecting citizenship with the classroom. Assessing and reporting on habits

If necessary, you might like to guide students to read over the notes and discuss the roles and language required of a chairperson or secretary to prepare them for the activity9.

● tracking students' progress in the use of thinking routines and in the development of their writing ability using a variety.. of formative assessment tools

Microphone and 600 ohm line conduits shall be mechanically and electrically connected to receptacle boxes and electrically grounded to the audio system ground point.. Lines in

– Lower of cost/NRV, sales or return and weighted average cost of inventory costing

Because simultaneous localization, mapping and moving object tracking is a more general process based on the integration of SLAM and moving object tracking, it inherits the