Research Article A Hybrid Cluster-Based Target Tracking...

Hindawi Publishing CorporationInternational Journal of Distributed Sensor NetworksVolume 2013 Article ID 494863 16 pageshttpdxdoiorg1011552013494863

Research ArticleA Hybrid Cluster-Based Target Tracking Protocol for WirelessSensor Networks

Zhibo Wang1 Wei Lou23 Zhi Wang1 Junchao Ma3 and Honglong Chen4

1 State Key Laboratory of Industrial Control Technology Zhejiang University Hangzhou 310027 China2 Shenzhen Research Institute The Hong Kong Polytechnic University Shenzhen 518057 China3Department of Computing the Hong Kong Polytechnic University Hong Kong4College of Information and Control Engineering China University of Petroleum Qingdao 266580 China

Correspondence should be addressed to Zhi Wang wangzhiiipczjueducn

Received 11 November 2012 Revised 15 February 2013 Accepted 7 March 2013

Academic Editor Tian He

Copyright copy 2013 Zhibo Wang et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Target tracking is a typical and important application of wireless sensor networks (WSNs) In consideration of the networkscalability and energy efficiency for target tracking in large-scaleWSNs it has been employed as an effective solution by organizingthe WSNs into clusters However tracking a moving target in cluster-based WSNs suffers a boundary problem when the targetmoves across or along the boundaries of clusters as the static clustermembership prevents sensors in different clusters from sharinginformation In this paper we propose a novel mobility management protocol called hybrid cluster-based target tracking (HCTT)which integrates on-demand dynamic clustering into a cluster-basedWSN for target tracking By constructing on-demand dynamicclusters at boundary regions nodes from different static clusters that detect the target can temporarily share information and thetracking task can be handed over smoothly from one static cluster to another As the target moves static clusters and on-demanddynamic clusters alternately manage the target tracking task Simulation results show that the proposed protocol performs betterin tracking the moving target when compared with other typical target tracking protocols

1 Introduction

Target tracking is considered important in WSNs as itis a base for many practical applications such as battle-field surveillance emergency rescue disaster response andpatient monitoring [1] Generally speaking target trackingaims to detect the presence of a target and compute reliableestimates of its locations while the target moves within anarea of interest and forward these estimates to the base stationin a timely manner

It is known that the cluster structure can provide benefitsfor large-scale WSNs For example it facilitates spatial reuseof resources to increase the system capacity [2] it also benefitslocal collaboration and routing [3 4] Recently the clusterstructure is gradually adopted for solving the target trackingproblem [5ndash13] Normally nodes surrounding the targetcollaborate with each other to estimate the location of thetarget In [5ndash10] dynamic clustering approaches are used thatdynamically wake up a group of nodes to construct a cluster

for local collaboration when the target moves into a regionClusters are constructed dynamically as the target movesDynamic clustering is obviously an efficient way for localsensor collaborations because clusters formed at each timeinstant change dynamically as the target moves Howeverthe dynamic clustering process repeating as the targetmovesis energy costly as it incurs much overhead for formingand dismissing clusters Besides dynamic clustering does notconsider how to efficiently send data to the sink which isanother important aspect of target tracking

In contrast the target tracking in [11ndash13] uses the staticcluster for the network scalability and energy efficiency (theterm of ldquostatic clusterrdquo does not mean that the cluster willnot change during the networkrsquos entire lifetime It meansthat the cluster structure will keep unchanged for a relativelylonger period of time compared to a temporally formeddynamic cluster until the next round of clustering processbegins In this way as shown in the LEACH protocol [14]each sensor node has the probability of becoming a cluster

2 International Journal of Distributed Sensor Networks

head so as to balance the energy load) It uses a predictivemechanism to inform cluster heads about the approachingtarget and then the corresponding cluster head wakes upa number of appropriate nodes right before the arrival ofthe target The overhead can be saved as the tracking taskis handed over from one static cluster to another with-out costly dynamic clustering processes It also provides ascalable structure for coordinating and managing networksTherefore static cluster-based approaches are more suitablefor target tracking in large-scale sensor networks Howeverthe static cluster membership prevents sensors in differentclusters from collaborating and sharing information witheach other which causes a so-called boundary problem whenthe target moves across or along the boundaries of clustersThe boundary problem will result in the increase of trackinguncertainty or even the loss of the target Therefore a newprotocol is required to solve the boundary problem and realizethe tradeoff between energy consumption and local sensorcollaboration for cluster-based sensor networks

In this paper we propose a novel distributed mobil-ity management protocol called hybrid cluster-based targettracking (HCTT) for efficient target tracking in a large-scalecluster-based WSN HCTT integrates on-demand dynamicclustering into a scalable cluster-based WSN with the helpof boundary nodes which facilitates sensorsrsquo collaborationamong clusters to solve the boundary problem As shown inFigure 1 when the target is inside a static cluster the clusteris responsible for target tracking as the target moves close tothe boundaries of clusters an on-demand dynamic clusteringprocess will be triggered to manage the tracking task so asto avoid the boundary problem The on-demand dynamiccluster will dismiss soon after the target moves away fromthe boundaries As shown in Figure 1 the sequential clustersfor tracking the target are 119860 rarr 1198631 rarr 119862 rarr 1198632 rarr

119864 When the target moves static clusters and on-demanddynamic clusters alternately manage the tracking task Byintegrating on-demand dynamic clustering into a scalablecluster-based structure nodes belonging to different staticclusters can share information which guarantees smoothlytarget tracking and realizes well tradeoff between energyconsumption and local sensor collaboration Note that aconference paper [15] containing some preliminary resultsof this paper appeared at IEEE DCOSS 2010 This paper hasbeen revised by following the comments of the reviewersand the feedback of other researchers Moreover we analyzethe computation and communication complexity of HCTTand extend the performance evaluation to multihop clusterscenario as opposed to just one-hop scenario

The main contributions of this paper are summarized asfollows

(i) We address the boundary problem for target trackingin a cluster-based WSN

(ii) We present a novel hybrid cluster-based target track-ing protocol which balances well between the energyconsumption and local sensor collaboration

(iii) The proposed algorithm is scalable to multi-hopstatic clusters and solves the problem of trackinguncertainty due to prediction errors

Dynamic cluster

Static cluster

Cluster headMember nodeLeader of dynamic cluster

119861

119860

119862

1198631

1198632

119864

119865

119863

Figure 1 Illustration of HCTT for target tracking in a cluster-basedWSN

(iv) We conduct simulations to show the efficiency of theproposed scheme compared with other typical targettracking solutions

The rest of the paper is organized as follows Section 2reviews the related work about target tracking in WSNs Sec-tion 3 describes the systemmodel and the boundary problemfor target tracking in a cluster-based WSN In Section 4we present the hybrid cluster-based target tracking protocolin detail The analysis of HCTT is presented in Section 5Section 6 demonstrates the performance evaluation for one-hop clusters and multi-hop clusters Finally we conclude thepaper in Section 7

2 Related Work

Target tracking in WSNs has received considerable atten-tion from various angles and a lot of protocols havebeen proposed Zhao et al proposed an information-drivendynamic sensor collaboration mechanism for target tracking[16] Brooks et al presented a distributed entity-trackingframework for sensor networks [17] Chen et al proposeddistributed sensor scheduling algorithms for binary sensornetworks [18] and acoustic sensor networks [19] respectivelyWang et al proposed a novel localization algorithm fortarget tracking [20] Vigilnet an energy-efficient and real-time integrated system for target tracking was designed andimplemented in [21 22] A distributed TDMA schedulingalgorithm for target tracking in ultrasonic sensor networkswas proposed in [23] In [24] the authors proposed a securelocalization scheme to defend against some typical attacksthat may affect target tracking accuracy More target trackingprotocols can be found in [25 26]

To balance energy consumption and sensor collaborationa lot of dynamic clustering protocols have been proposedfor target tracking in WSNs Yang et al proposed anadaptive dynamic cluster-based tracking (ADCT) protocol

International Journal of Distributed Sensor Networks 3

that dynamically selects cluster heads and wakes up nodesto construct clusters with the help of a prediction algo-rithm as the target moves in the network [6] Zhang andCao proposed a dynamic convoy tree-based collaboration(DCTC) framework to detect and track the mobile target[7] The convoy tree can be considered as a cluster which isdynamically configured by adding and pruning some nodesas the target moves Ji et al proposed a dynamic cluster-based structure for object detection and tracking [8] Jinet al also proposed a dynamic clustering mechanism fortarget tracking in WSNs that balances the missing rate andenergy consumption [9] Medeiros et al proposed a dynamicclustering algorithm for target tracking in wireless cameraNetworks [10] A decentralized dynamic clustering protocolfor acoustic target tracking which relies on a static backboneof sparsely placed high-capacity sensors was proposed in [5]As mentioned in Section 1 dynamic clustering is efficient forlocal sensor collaboration but incurs too much overhead forcluster construction and dismiss as the target moves In thispaper HCTT overcomes the clustering overhead by relyingon a preexisting static cluster structure

Several target tracking protocols are based on static clus-ter structure [11ndash13] Sensor nodes are organized into clustersby using suitable clustering protocols such as LEACH [14]and HEED [27] With the cluster structure Yang and Sikdarproposed a distributed predictive tracking (DPT) protocolthat predicts the next location of the target and informs thecluster head about the approaching targetThe correspondingcluster head then wakes up the closet three sensors nodesaround the predicted location before the arrival of the target[11] Wang et al [12] proposed a hierarchical predictionstrategy (HPS) that also relies on cluster structure for targettracking and implemented a real target tracking system in[13] Compared with dynamic clustering protocols cluster-based target tracking protocols take the advantages of under-lying cluster structure which is especially suitable for targettracking in large-scale networks However the static clustermembership prevents sensors in different clusters from col-laborating and sharing information with each other whichcauses the boundary problemwhen the targetmoves across oralong the boundaries of clusters In this paper HCTT solvesthe boundary problem by integrating on-demand dynamicclustering into the scalable cluster structure

3 System Model and Problem Statement

In this section we first present the systemmodel and thenwedescribe the boundary problem for target tracking in cluster-based WSNs

31 SystemModel Weassume that a large-scaleWSNconsist-ing of 119899 static sensor nodes is deployed in a two-dimensionalarea of interest for detecting a single moving target The sinknode is deployed at the center of the network We assumethat sensor nodes are randomly deployed following a Poissondistribution with a density of 120582 That is given an area 119860 the

number of sensor nodes in the area119873(119860) follows a Poissondistribution with parameter 120582119860 as follows

119875119903 (119873 (119860) = 119896) =(120582119860)119896

119896sdot (119890minus120582119860

) 119896 = 0 1 infin (1)

We assume that each sensor node has an identicalcommunication range 119903119888 The sensor nodes within the com-munication range of a sensor node are called its neighbornodes That is 119873(V119894) = V119895 | 119889(119897119894 119897119895) le 119903119888 is the set ofneighbor nodes of sensor node V119894 where 119897119894 is the location ofV119894 119897119895 is the location of V119895 and 119889(119897119894 119897119895) is the Euclidean distancebetween V119894 and V119895

In general the received signal of a sensor node about thetarget reduces with the increase of the distance between thesensor node and the target An attenuated disk sensingmodelis adopted to capture the sensing quality as follows

119903119894 =

120573

119889120572 (119897119894 119871) 119889 (119897119894 119871) le 119903119904

0 (119897119894 119871) gt 119903119904

(2)

where 119903119894 is the received signal of sensor node V119894 120573 is theoriginal strength of emitted signal of the target 120572 is the pathattenuation exponent 119903119904 is the sensing range 119897119894 is the locationof V119894 119871 is the target location and 119889(119897119894 119871) is the Euclideandistance between sensor node V119894 and the target

Note that the sensing region of a sensor node V119894 denotedby 119877(V119894 119903119904) is a disk with center 119897119894 and radius 119903119904 A target willbe detected by the sensor V119894 when it appears in the sensingregion 119877(V119894 119903119904) Conversely only sensor nodes within thedistance of 119903119904 of the target can detect the target Thereforewe call the disk centered at the target and with radius 119903119904 asthe monitoring region of the target as any sensor within thisregion can monitor the target

Each node can operate in three states It can transmitpackets receive packets and sense the target in active state Insensing state it can only perform sensing operation In sleepstate it sleeps for most of time and wakes up periodically tosense the target and listen to messages Since communicationoperation dominates the energy consumption we mainlyconsider the energy consumption for communication in thispaper

We assume that the WSN is organized as 119898 clusters byusing any suitable clustering algorithm in [28] In this paperwe use the LEACH [14] to organize nodes into static clustersTheLEACHprotocol is an energy efficient adaptive clusteringprotocol which is distributed and energy balanced Eachcluster 119894 has 119899119894 nodes including one cluster head and 119899119894 minus 1

1-hop members Each node is aware of its own location andsome local information of its neighbors but has no globaltopology information To fulfill the monitoring task clusterheads are responsible for selecting somemembers tomonitorthe target and deal with the handoff process when the targetmoves

To ease the description of the protocol we adopt thefollowing notations throughout the paper

(i) 119899 the number of sensor nodes(ii) 119898 the number of disjoint static clusters


Monitoring region

119861

119889

119888

119887

119886

119891119890

119860

119862

CH

CH

CH

(a)

Monitoring region

Actual location

Predicted location

119861

119889

119888

119887

119886

119891119890

119860

119862

CH

CH

CH

(b)

Figure 2 The boundary problem in a cluster-based WSN (a) high localization uncertainty due to insufficient active nodes (b) loss of targetdue to incorrect prediction of the target location

(iii) 119903119888 the communication range of each node(iv) 119903119904 the sensing range of each node(v) 119877(V119894 119903119904) the sensing region of node V119894 with sensing

radius 119903119904(vi) 119897119894 the location of node V119894 119897119894 = (119909119894 119910119894) where (119909119894 119910119894)

are the coordinates of the node(vii) 119871(119905) the location of the target at time 119905(viii) 119866 the wireless sensor network 119866 = (119881 119864)

(ix) 119881 the set of all sensor nodes 119881 = V1 V2 V119899

(x) 119864 the set of edges referred to all connected links ofnodes 119864 = 119890 = (V119894 V119895) | V119894 V119895 sube 119881 and 119889(119897119894 119897119895) le

119903119888 and 119894 = 119895

(xi) 119873(V119894) the neighbor set of node V119894 119873(V119894) = V119895 |

119889(119897119894 119897119895) le 119903119888

(xii) 119862119894 the set of nodes in cluster 119894 119862119894 sube 119881 Note that⋃119898

119894=1119862119894 = 119881 and 119862119894⋂119862119895 = 120601 where 119894 = 119895 | 119894 119895 =

1 2 119898

(xiii) 119862(V119894) the cluster which node V119894 belongs to(xiv) 119867 the set of all cluster heads 119867 =

V1198671

V1198672

V119867119898

sub 119881 where V119867119894

denotes thecluster head of cluster 119894

32 Boundary Problem When tracking a target in a surveil-lance area multiple nodes surrounding the target collaborateto make the collected information more complete reliableand accurate There is no problem when the target is insidea cluster as all activated sensors belong to the same clusterand they can communicate effectively However when thetarget moves across or along the boundaries of multipleclusters the boundary problem occursThat is the local node

collaboration becomes incomplete and unreliable becausesensor nodes that can monitor the target belong to differentclusters which increases the uncertainty of the localizationof the target or even results in the loss of the target due to theinsufficient sensing reports or the incorrect prediction of thetargetrsquos location

Figure 2 shows two cases of the boundary problem fortarget tracking in a cluster-basedWSN As the target is insidecluster 119860 cluster 119860 is activated to track the target whereasclusters 119861 and 119862 are in the sleep state for energy savingpurpose For the case in Figure 2(a) when the target movesclose to the boundaries of clusters nodes 119886 119887 119888 119889 119890 and119891 which can monitor the target belong to three differentclusters 119860 119861 and 119862 Only nodes 119886 and 119887 can sense the targetas they are active while other nodes cannot as they are in thesleep state The network cannot successfully locate the targetdue to insufficient informationwhichmay affect the accuracyof predicting the targetrsquos next position or even result in theloss of the target For the case shown in Figure 2(b) the nextlocation of the target is predicted to be in cluster 119861 whereasthe target actually moves into cluster 119862 Nodes in cluster 119861are activated in advance according to the predicted locationof the targetHowever none of them can sense the target sincethe target moves into cluster 119862 Therefore prediction errormay also result in the loss of the target

In this paper we propose a hybrid cluster-based targettracking protocol (HCTT) for efficient target tracking in acluster-based WSN HCTT integrates on-demand dynamicclustering into a scalable cluster-based structure with thehelp of boundary nodes That is when the target movesinside a cluster the static cluster is responsible for local nodecollaboration and target tracking when the target approachesthe boundaries of clusters a dynamic clustering process willbe triggered to form a dynamic cluster to solve the boundary


Sensor networkdeployment

Static clusterformation

Boundary nodeformation

Target detection Static cluster managetarget tracking task

Dynamicclustering

D2SICHandoff

S2DIC

HandoffD2DIC

Handoff

HCTT

Dynamic cluster managetarget tracking task

Figure 3 Overview of hybrid cluster-based target tracking protocol

problem The on-demand dynamic cluster will disappearsoon after the target moves away from the boundaries Asthe target moves in the network static clusters and on-demand dynamic clusters alternately handle the tracking taskeffectively

4 Hybrid Cluster-Based TargetTracking Protocol (HCTT)

In this section we first give a high-level overview of HCTTand then describe the protocol in detail Figure 3 outlinesthe overview of the system and illustrates the flowchart ofHCTT After being deployed sensor nodes are organized intostatic clusters according to any suitable clustering algorithmBoundary nodes of each cluster are also formed When atarget is in the network a static cluster sensing the targetwakes up to track the target When the target approaches theboundary the boundary nodes can detect the target and anon-demand dynamic cluster will be constructed in advancefor smoothly tracking the target before the target reaches theboundary As the target moves across the boundaries staticclusters and on-demand dynamic clusters alternately managethe tracking task Different types of intercluster handoff (iethe S2DIC handoff from a static cluster to a dynamic clusterthe D2SIC handoff from a dynamic cluster to a static clusterand the D2DIC handoff from a former dynamic cluster to anew dynamic cluster) take place between any two sequentialclusters Moreover the dynamic cluster will dismiss after thetarget moves away from the boundary and enters anothercluster With the help of boundary nodes HCTT integrateson-demand dynamic clustering into a scalable cluster-basedWSN for target tracking which facilitates the sensorsrsquo collab-oration among clusters and solves the boundary problem

In the following we elaborate on the design and imple-mentation of three major components of HCTT includingthe boundary node formation dynamic clustering and inter-cluster handoff

41 Boundary Node Formation In HCTT a dynamic clusterwill be constructed when the target approaches the bound-aries of multiple clusters A challenging issue is how thesystem finds the scenario when the target is approachingthe boundaries especially in a fully distributed way Assensor nodes are randomly deployed static clusters areusually formed irregularly It is difficult or even impossible

to calculate the geometrical boundaries of irregular clustersIn this paper we use boundary nodes to solve this issue in afully distributed way

Definition 1 Boundary node of a static cluster A node V119894 isdefined as a boundary node of its static cluster if there existsat least one of its neighbor nodes V119895 such that 119889(119897119894 119897119895) le 119903119904

and 119862(V119894) = 119862(V119895)

In contrast a node is defined as an internal node of itsstatic cluster if it is not a boundary node As each node isaware of its own location and its neighbor information itchecks its neighbor list to determine whether there exists anode belonging to another cluster within its sensing range Ifyes it is a boundary node otherwise it is an internal nodeThe boundary node set of a cluster 119862119894 denoted by 119861(119862119894) isformed by all the boundary nodes in 119862119894 The internal node setof a cluster 119862119894 denoted by 119868(119862119894) is formed by all the internalnodes in 119862119894 The pseudocodes of boundary node formationprocess are described in Algorithm 1

After boundary nodes are identified each cluster can bepartitioned into three parts safety region boundary regionand alert region

Safety Region The safety region of a cluster 119862119894 denoted by119877119878(119862119894) is the region in cluster 119862119894 that can be monitored by atleast one internal node of 119862119894 but not by any boundary nodeof 119862119894 That is 119877119878(119862119894) can be formulated as

119877119878 (119862119894) = ⋃

forallV119894isin119868(119862119894)

119877 (V119894 119903119904) minus ⋃

forallV119894isin119861(119862119894)

119877 (V119894 119903119904) (3)

Boundary Region The boundary region of a cluster 119862119894denoted by 119877119861(119862119894) is the region that can be monitored bysome boundary nodes of both 119862119894 and any of its adjacentclusters at the same time That is 119877119861(119862119894) can be formulatedas

119877119861 (119862119894) = ⋃

forallV119894isin119861(119862119894)forallV119895isin119861(119862119895)forall119862119895 = 119862119894

(119877 (V119894 119903119904) cap 119877 (V119895 119903119904))

(4)

Alert Region The alert region of a cluster 119862119894 denoted by119877119860(119862119894) is the region that can be monitored by at least one


Input Graph G = 119881 119864 cluster sets 119862119894 119894 = 1 2 119898Output 119861(119862

119894) 119894 = 1 2 119898

(1) for each cluster 119862119894 do(2) 119861(119862119894) larr 120601

(3) for each node V119895 isin 119862119894 do(4) while existV119896 isin 119873(V119895) such that 119889(119897119896 119897119895) le 119903119904 and

119862(V119895) = 119862(V119896) do(5) V119895 sdot 119904119905119886119905119890 larr 119887119900119906119899119889119886119903119910

(6) 119861 (119862119894) larr 119861 (119862119894) cup V119895

Algorithm 1 Boundary node formation

Alert region

Boundary region

Internal nodeBoundary nodeCluster head

Safety region

Figure 4 Demonstration of safety region alert region and bound-ary region for a circular cluster

boundary node of 119862119894 but not belongs to the boundary regionof 119862119894 That is 119877119860(119862119894) can be formulated as

119877119860 (119862119894) = ⋃

forallV119894isin119861(119862119894)

119877 (V119894 119903119904) minus 119877119861 (119862119894) (5)

Figure 4 illustrates the three different regions for a circularcluster The white region denotes the safety region the lightgray region denotes the alert region and the dark gray regiondenotes the boundary region We can observe that if thetarget moves from the safety region into the boundary regionthrough the alert region nodes belonging to different clusterswill detect the target More specifically we have the followinglemmas

Lemma 2 All boundary nodes lie in the boundary region

Proof For any boundary node V119894 of cluster 119862119894 by definitionthere exists a neighbor node V119895 of cluster 119862119895 such that119889(119897119894 119897119895) le 119903119904 and 119862(V119894) = 119862(V119895) That is V119895 is also a boundarynode of cluster 119862119895 Since 119889(119897119894 119897119895) le 119903119904 we can get V119894 lies in119877(V119894 119903119904) cap 119877(V119895 119903119904)

Lemma 3 If nodes detecting the target are all internal nodesof a cluster the target moves into the safety region of the clusterand vice versa

Proof If nodes detecting the target are all internal nodes of acluster which suggests that the target cannot be detected byany boundary node According to (3) the target moves intothe safety region It is also true that if the target moves intothe safety region of a cluster obviously all nodes detecting thetarget are internal nodes of the cluster

Lemma 4 If nodes detecting the target belong to differentclusters the targetmoves into the boundary region of one clusterand vice versa

Lemma 5 If nodes detecting the target all belong to one clusterand there is at least one boundary node among them the targetmoves into the alert region of the cluster and vice versa

The proofs of Lemmas 4 and 5 are similar to that ofLemma 3

Based on Lemmas 3 4 and 5 a cluster head can decidewhich region the target locates when it receives data reportsfrom sensor nodes

42 Dynamic Clustering Dynamic clustering is triggered ondemand in order to solve the boundary problem in a cluster-based WSN There are two cases in which a dynamic clusteris needed The first case is shown in Figure 5(a) where thecurrent active cluster is cluster 119860 When the target movesfrom point 119886 to point 119888 via point 119887 a dynamic cluster1198631 willbe constructed for smoothly tracking the target The secondcase is shown in the Figure 5(b) where the current activecluster is a dynamic cluster1198632 When the target moves alongthe boundaries from point 119898 to point 119902 via points 119900 and119901 using only one dynamic cluster cannot satisfy the targettracking requirement In this case another dynamic cluster1198633will be constructed when the target moves from point 119900 topoint119901 Note that clusters of square shapes in Figure 5 are justused to help explain the handoff problem in target trackingThe clusters might not have any regular shape in practice

Though the dynamic cluster must be constructed inadvance before the target moves across the boundaries ofclusters it is costly if the dynamic cluster is constructed tooearly which not only wastes nodesrsquo energy consumption but


1198631

Alert regionBoundary regionSafety region

119860

119861

119862119863

119886

119887119888

119889

119890

(a)


119860

119861

119862

119863

1198633

1198632119898

119899

119902

119901119900

(b)

Figure 5 Dynamic clustering and handoff between clusters (a) one dynamic cluster (b) two consecutive dynamic clusters

also may become useless quickly As stated in Lemma 5 ifthe target is detected by any boundary node the target movesinto the alert region which means that the target is closelyapproaching the boundary region It is reasonable to triggerthe on-demand dynamic clustering process when the targetmoves into the alert region that is when there is at least oneboundary node detecting the target The dynamic clusteringprocess consists of three phases leader selection dynamiccluster construction and boundary node formation

421 Leader Selection The first step to construct a dynamiccluster is to elect a cluster head from sensor nodes Thedynamic clustering process should be triggered once thetarget moves into the boundary region Therefore the firstboundary node detecting the target is supposed to be thecluster head and construct a dynamic cluster Howeversensor nodes may detect the target at the same time so thesenodes may compete to be the cluster head Therefore analgorithmmust be proposed to guarantee only one boundarynode to be the cluster head

Once a boundary node detects a target if it does notreceive any election message from its neighbor nodes itbroadcasts an electionmessage to its one-hop neighbor nodesIf a boundary node receives some election messages beforedetecting the target it would not broadcast any messageThe election message includes the ID the received signal ofthe target and the residual energy of the sensor node Afterit broadcasts the election message it waits for a predefinedtimeout period to receive the election message from othersensor nodes A sensor node may or may not receive theelection message during the timeout period otherwise ifit does not receive any election message after the timeoutperiod it will elect itself as the cluster head If it receivesmultiple electionmessages from its neighbors the node withthe highest received signal will be elected as the cluster headIn case somenodes have the same received signal the residualenergy can be used to break the tie

422 Dynamic Cluster Construction After a node is electedas the cluster it broadcasts a recruit message asking itsneighbor nodes to join in the cluster Each node receivingthe recruit message establishes a new table containing theinformation of the dynamic cluster for example the ID of theleader and the working state of the dynamic cluster and thenreplies a confirm message to the leader Upon receiving theconfirm message from a node the leader puts the node intoits member list

Note that once a dynamic cluster is constructed a nodemay belong to a static cluster and a dynamic cluster at thesame time When the node generates some data it needsto decide which cluster head it should report to This isimportant for the inter-cluster handoff procedure andwewilldescribe the details in Section 43

423 Boundary Node Formation Boundary nodes are alsonecessary for dynamic clusters It can inform the cluster headof the dynamic cluster in advance if the target is about to leavethe dynamic cluster

Definition 6 Boundary node of a dynamic cluster A node V119894is defined as a boundary node of a dynamic cluster119863 if thereexists at least one neighbor node V119895 such that V119895 notin 119863 and119889(119897119894 119897119895) le 119903119904

The way of forming boundary nodes of a dynamic clusteris different from that of a static cluster Each node V119894 checksits neighbor list to see if there exists a neighbor node thatdoes not belong to the dynamic cluster and also is within itssensing range If such a neighbor node can be found V119894 isa boundary node of the dynamic cluster otherwise it is aninternal node of the dynamic cluster Once boundary nodesof the dynamic cluster are identified the dynamic clusteringprocess is completely finished After the dynamic cluster isconstructed it stays in the waiting state for the coming of thetargetThepseudocodes of the dynamic clustering process aredescribed in Algorithm 2


Phase I Leader selection(1) Leader node 119871dc larr 120601

(2) for each sensor node detecting the target do(3) if it does not receive any electionmessage then(4) broadcasts its electionmessage and waits for a predefined

timeout period for election messages from neighbor nodes(5) if it does not receive any message during the period

then(6) elects itself as the cluster head 119871dc(7) else(8) elects the node with the highest received signal as

the cluster head 119871dcPhase II Dynamic cluster construction(1) dynamic cluster set 119862dc larr 120601

(2) 119871dc broadcasts a recruit message(3) while node V119896 receiving the recruit message do(4) V119896leaderlarr 119871dc(5) V119896 replies a confirmmessage(6) while 119871dc receives a confirmmessage from V119896 do(7) 119862dc larr 119862dc cup V119896

Phase III Boundary node formation(1) boundary nodes set 119861dc larr 120601

(2) for each node V119896 isin 119862dc do(3) while existV119901 isin 119873(V119896) such that 119889(119897119896 119897119901) le 119903119904 and

V119901119897119890119886119889119890119903 = 119871dc do(4) V119896statelarr boundary(5) 119861dc larr 119861dc cup V119896

Algorithm 2 Dynamic clustering

Table 1 Message reporting priority

S D Reported cluster headActive Idle CH119904Sleep Idle CH119889Sleep Active CH119889

43 Intercluster Handoff Before introducing the details ofinter-cluster handoffprocess we first give themessage report-ing priority order used in our protocol

Message Reporting Priority Each cluster can work at one ofthree states active idle (waiting) and sleep The active statehas the highest priority formessage reporting while the sleepstate has the lowest priority If a node belongs to a static clusterand a dynamic cluster simultaneously it always reports to thecluster head with higher priority One exception case is thatif the node is a boundary node of a static cluster it reports tothe headers of both clusters Table 1 illustrates different datareporting scenarios where a node belongs to a static clusterand a dynamic cluster at the same time Here 119878 and119863 denotea static cluster and a dynamic cluster respectively and 119862119867119904

and 119862119867119889 denote the cluster heads of 119878 and119863 respectivelyGenerally the tracking task should be handled by the

most suitable cluster As the target moves in the networkthe task should be handed over from the current clusterto another more suitable one which is taken charge of by

the inter-cluster handoff process The inter-cluster handoffoccurs in three typical scenarios handoff from a static clusterto a dynamic cluster (eg from cluster 119860 to cluster 1198631 asshown in Figure 5(a)) handoff from a dynamic cluster to astatic cluster (eg from1198631 to119863 as shown in Figure 5(a)) andhandoff from a dynamic cluster to another dynamic cluster(eg from1198632 to1198633 as shown in Figure 5(b)) In the followingpart wewill describe how to efficiently implement these threetypes of handoff processes

431 S2DIC Handoff When the target locates within acluster it is tracked by the current active static cluster As thetarget moves into the alert region boundary nodes can detectthe target Upon detecting the target by a boundary node adynamic cluster is constructed in advance for the coming ofthe target However the handoff should not take place rightaway once the new dynamic cluster is constructed since themovement of the target is uncertain The target may movetowards the boundary but it may also turn around and moveaway from the boundary As shown in Figure 6 when thetarget moves to point 119887 in the alert region a dynamic cluster1198631 is constructed before the target moves into the boundaryregion If the targetrsquos trajectory is T1 the dynamic clusteris more suitable than the active static cluster for trackingthe target However the target may follow trajectories T2or T3 For these trajectories it is inappropriate to performthe handoff since the target does not actually move intothe boundary region Especially for the trajectory of T3


119860

119886

119888

119887

11987931198792

1198791

1198634119861

Boundary regionSafety region Alert region

Figure 6 Different moving trajectories of the target

unnecessary dynamic clustering and handoff may take placefrequently if we do not have an effective handoff mechanism

To minimize unnecessary dynamic clustering and hand-off the S2DIC handoff starts when the target is confirmedto be in the boundary region When the dynamic cluster isconstructed some boundary nodes of the current active staticcluster join into the dynamic cluster These nodes are in thewaiting state while they keep detecting the target When anyof these nodes detects the target it sends the sensing reportto the cluster head of the dynamic cluster Once this clusterhead receives the sensing report the target is confirmed to bein the boundary region and the handoff process starts

The S2DIC handoff procedure is shown in Figure 7(a) thedynamic cluster head say 119871dc first sends a request messageto the active static cluster head 119862119867119894 to ask for the handoffof the leadership 119862119867119894 replies a ℎ119886119899119889119900119891119891message containingthe historical estimations of the targetrsquos locations and handsthe leadership over to 119871dc Once 119871dc receives the ℎ119886119899119889119900119891119891message it first broadcasts a work message to activate all itsmember nodes and then replies an 119886119888119896119899119900119908119897119890119889119892119890 message to119862119867119894 After receiving the 119886119888119896119899119900119908119897119890119889119892119890 message 119862119867119894 broad-casts a 119904119897119890119890119901 message to its member nodes Each membernode not belonging to the active dynamic cluster goes intothe sleep state to save energy consumption

If the dynamic cluster is not suitable for tracking thetarget it will be dismissed As shown in Figure 6 whenthe target moves to point 119888 following the trajectory 1198792 thedynamic cluster 1198634 is not effective anymore and will bedismissed The dynamic cluster dismissal procedure worksas follows for a dynamic cluster in idle state if the clusterhead of the dynamic cluster receives sensing reports fromthe boundary nodes of the dynamic cluster it confirms thatthe dynamic cluster is not needed anymore Then the clusterhead of the dynamic cluster sends a resignmessage to informthe cluster head of the active static cluster After gettingthe acknowledged message from the cluster head of theactive static cluster it broadcasts a dismiss message to all itsmembers Each node when receiving the dismiss messagequits the dynamic cluster and deletes the information tablebuilt for the dynamic cluster

432 D2SIC Handoff When a dynamic cluster is currentlyhandling the tracking task nodes sensing the target send their

sensing reports to the cluster head of active dynamic clusterIf none of these sensing nodes is a boundary node of anystatic cluster the target has moved away from the boundaryregion and entered the safety region of a static cluster thenthe D2SIC handoff process is triggered

The procedure of D2SIC handoff is shown in Figure 7(b)the active cluster head 119871dc sends a handoff message to thecluster head of next static cluster 119862119867119895 Once 119862119867119895 receives ahandoff message it broadcasts a119908119900119903119896message to wake up itsmember nodes then replies an acknowledge message to 119871dcAfter receiving the acknowledge message 119871dc broadcasts adismissmessage Each node when receiving the dismissmes-sage quits the dynamic cluster and deletes the informationtable for the dynamic cluster

433 D2DIC Handoff When a dynamic cluster is currentlyactive for the tracking task the handoff condition for theD2SIC handoff may not be met as the target moves intothe boundary region As shown in Figure 5(b) one dynamiccluster1198632 cannot track the target all the time when the targetmoves along the boundaries of static clusters In this scenariothe D2DIC handoff can construct another new dynamiccluster 1198633 to continue the tracking task if some boundarynodes of the active dynamic cluster detect the target Notethat the D2SIC handoff has a higher priority than the D2DIChandoff as the latter onewill causemore cost and longer delaythan the former one

The new dynamic cluster is generally a more preferablechoice than the previous one for tracking the target as thetarget is at the center of the new dynamic cluster while it isat the boundary of the previous dynamic cluster Thereforethe handoff takes place immediately after the new dynamiccluster is constructed The D2DIC handoff procedure isshown in Figure 7(c) the new cluster head119871dc1015840 sends a requestmessage to ask for the handoff of the leadership The currentactive cluster head 119871dc replies a ℎ119886119899119889119900119891119891message containingthe historical estimations of the targetrsquos location to handover the leadership to the cluster head of the new dynamiccluster After receiving the leadership 119871dc1015840 broadcasts a workmessage to inform its members to be responsible for trackingthe target Each node detecting the target sends the sensingreports to 119871dc1015840 Then 119871dc1015840 replies an acknowledge messageto the 119871dc After receiving the 119886119888119896119899119900119908119897119890119889119892119890 message 119871dcbroadcasts a 119889119894119904119898119894119904119904message Each node when receiving thedismiss message quits from the dynamic cluster and deletesthe information table for the dynamic cluster

5 Analysis of HCTT

In this section we will present the computation complexityand communication complexity for proposed HCTT

As mentioned before we have assumed that 119899 static sen-sor nodes are randomly deployed in the area of interest Thedeployment of these nodes follows the Poisson distributionwith density of 120582 Therefore the average number of neighbornodes of one node is 1205821205871199032

119888minus 1


Request

Handoff

AcknowledgeWork

Sleep

CH119894 119871dc

(a)

Handoff

AcknowledgeWork

Dismiss

CH119895119871dc

(b)

Request

Handoff

AcknowledgeWork

Dismiss

119871dc998400119871dc

(c)

Figure 7 Illustration of intercluster handoffprocesses (a) S2DIChandoffprocess (b)D2SIChandoffprocess and (c)D2DIChandoffprocess

Theorem 7 The computation complexity and communicationcomplexity of the boundary node formation process are both119874(119899)

Proof In the boundary node formation process each nodebroadcasts amessage to its one-hop neighbors so the numberof messages transmitted is proportional to the number ofnodes Therefore the communication complexity of theboundary node formation process is 119874(119899) As presented inAlgorithm 1 each node computes to judge whether it is aboundary node or not In the worst case each node needs tocheck all its neighbors to make the decision that is 1205821205871199032

119888minus 1

times are computed to judge whether it is a boundary node ornot For the whole network 119899lowast(1205821205871199032

119888minus1) times are computed

in the worst case to complete the process of boundary nodeformation The density 120582 and communication range 119903119888 areusually fixed after deployment so 1205821205871199032

119888minus1 can be considered

as a constant Therefore the computation complexity of theprocess of boundary node formation is also 119874(119899)

Theorem 8 The computation complexity and communicationcomplexity of the dynamic clustering process are both Θ(1)

Proof Dynamic clustering process consists of three phasesleader selection dynamic cluster construction and boundarynode formation As presented in Algorithm 2 all the mes-sages transmitted are constrained in a one-hop cluster thatis the number of messages transmitted is determined by 120582and 119903119888 which can be considered as a constant Thereforethe communication complexity is Θ(1) In the worst caseeach dynamic node needs to check all its neighbors to decidewhether it is a boundary node of the dynamic cluster or notSince there are 1205821205871199032

119888nodes for each dynamic cluster the over-

all computation overhead in the worst case is (1205821205871199032119888)lowast(120582120587119903

2

119888minus

1) which also can be considered as a constant Therefore thecomputation complexity of the dynamic clustering process isΘ(1) We can conclude that the dynamic clustering is a localprocess which is independent to the size of network

Theorem 9 The computation complexity and communicationcomplexity of the inter-cluster handoff process are both Θ(1)

Proof Inter-cluster handoff includes three schemes S2DIChandoff D2SIC handoff and D2DIC handoff Figure 7 showsthat inter-cluster handoff is a local process The messageoverhead is much smaller than 120582120587119903

2

119888 since the process only

requires some control packets During the handoff processeach node involved makes its decision locally to be active orgo to sleep In the worst case sensor nodes of two clusters areinvolvedThat is the computation overhead in the worst caseis 21205821205871199032

119888 which can be considered as a constant Therefore

the communication complexity and computation complexityof the inter-cluster handoff process are both Θ(1) We canconclude that the inter-cluster handoff is a local processwhich is independent to the size of network

Theorem 10 The computation complexity and communica-tion complexity of the hybrid cluster-based target tracking areboth 119874(119899)

Proof Hybrid cluster-based target tracking protocol con-sists of three major components boundary node formationdynamic clustering and inter-cluster handoff Since the com-putation complexity andmessage complexity of the boundarynode formation process are 119874(119899) the computation complex-ity and message complexity of HCTT are also 119874(119899)

Although the computation complexity and communica-tion complexity of HCTT are 119874(119899) we can see that HCTT isinsensitive to the size of network if we overlook the cost ofboundary node formation process We have mentioned thatas described in [14] the static cluster structurewill not changeuntil a new round of clustering process The period betweentwo sequential rounds of clustering processes is usually quitelong The process of boundary node formation only takesplace once for one static cluster structure Thus during eachperiod of two sequential rounds only dynamic clustering andinter-cluster handoff processes take place as targets move


of which the computation complexity and communicationcomplexity are bothΘ(1) Therefore in certain sense HCTTis scalable as the size of network increases

6 Performance Evaluation

In this section we evaluate the performance of our proposedscheme through simulations Since ourHCTT scheme is pro-posed to solve the boundary problem in a cluster-basedWSNwe mainly compare the performance of HCTT with that ofDPT [11] Moreover we further compare the performance ofHCTT with other two typical tracking protocols DCTC [7]and ADCT [6]

We use MATLAB to perform our simulations The simu-lation environment is configured as follows A total of 6000sensor nodes are randomly deployed in the area of 400 times

400m2 for monitoring a moving target Sensor nodes areorganized as static clusters according to the LEACHprotocolThe sensing range of each node 119903119904 is fixed at 10m Thetransmission range of each node 119903119888 is set to be at least twiceas large as the sensing range that is 119903119888119903119904 ge 2 The size ofa control message used for constructing dynamic clusters is10 bytes The size of a sensing report generated by each nodefor detecting the target is 40 bytes The movement of a targetfollows the random-way point model The target randomlychooses a destination and moves toward this destination at arandom speed in [5 10] ms On arriving at the destinationthe node pauses for a predetermined time and then repeatsthe processThe prediction error which refers to the distancedifference between the estimated location computed by thesystem and the real location of the target varies from 0 to 119903119904Moreover each sensor node adopts the energy consumptionmodel presented in [14] To transmit a 119896-bit packet with arange 119903119888 the consumed energy is119864 = 119864119890 sdot119896+120598sdot119896sdot119903

2

119888 Typically

119864119890 = 50 nJbit and 120598 = 10 pJbitm2 For all the simulationresults presented in this paper each data point is an averageof 100 experiments

We use the followingmetrics to evaluate the performanceof our proposed scheme

(i) number of dynamic clusters the number of dynamicclusters generated during the target tracking process

(ii) missing probability the probability of missing thetarget as the target moves in the network

(iii) sensing coverage the ratio of the number of nodessensing the target to the number of nodes within themonitoring region

(iv) energy consumption the energy consumed for trans-mitting control messages and sensing reports

Note that although different localization approachesaffect the performance of target tracking significantly theyare not the major concern of this paper Here we use themetric of the sensing coverage to indicate the accuracy levelof location estimates of the target since the accuracy oflocalization is related to the information collected fromnodessurrounding the target It is expected that all nodes detectingthe target send their sensing reports to generate reliable and

0 100 200 300 400

50

100

150

200

250

300

350

Figure 8 Snapshot of target tracking using HCTT

accurate estimates of the target However many of themmaynot be able to sense the target successfully as they are inthe sleep state due to the prediction error or other factorsTherefore larger sensing coverage usually means a betterestimate of the targetrsquos location

Figure 8 shows a snapshot of using our proposed HCTTscheme for target tracking Sensor nodes are organized asstatic clusters Each cluster has only one cluster head whichis denoted with the snowflake mark The yellow pointsdenote the boundary nodes which are close to the boundariesof static clusters The light green line denotes the movingtrajectory of the targetTheblue crossmarks denote the nodessensing the target in static clustersThe small red circle marksdenote the nodes sensing the target in dynamic clusterswhich are denoted by the large circles We can see thatdynamic clusters are triggered on-demand when the targetmoves to the boundaries of clusters All these three inter-cluster handoff procedures S2DIC handoff D2SIC handoffand D2DIC handoff are shown in this snapshot when thetarget moves along the boundaries of clusters

61 In One-Hop Static Clusters Scenario In this scenariosensor nodes are organized as one-hop static clusters whereeach member node can send messages directly to the clusterhead

611 Number of Dynamic Clusters Figure 9 shows therelationship between the number of dynamic clusters andthe ratio of the transmission range to the sensing range119903119888119903119904 We can observe that the number of dynamic clustersdrops quickly as the ratio 119903119888119903119904 increases That is the smallerthe ratio 119903119888119903119904 is the more the generated dynamic clustersare The reason is that increasing the transmission rangemeans increasing the size of static clusters The larger thesize of each static cluster is the lower the probability that thetarget encounters the clusterrsquos boundaries is The number ofdynamic clusters drops quickly when 119903119888119903119904 lt 4 and dropsslowly when 119903119888119903119904 gt 4 Consequently we select the turning


2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

Num

ber o

f dyn

amic

clus

ters

119903119888119903119904

Figure 9The number of dynamic clusters versus 119903119888119903119904 in the HCTTalgorithm

00

10

20

30

40

50

60

70

80

90

100

HCTTDPT

ADCTDCTC

02 04 06 08 1

Miss

pro

babi

lity

()

Prediction error (119903119904)

Figure 10 The missing probability versus the prediction error forall algorithms

point of 119903119888119903119904 = 4 for the performance evaluation in thefollowing simulations

612 Missing Probability Figure 10 shows the effects ofthe prediction error on the missing probabilities of fourapproaches We can see that the missing probability of DPTis much higher than any of other three ones Even when theprediction error equals to zero DPT still has the probabilityof missing the target This shows the effects of the boundaryproblem in cluster-based sensor networks Moreover themissing probability of DPT increases as the prediction errorincreases which implies that the boundary problem willbecome worse when the prediction error increasesThe sametrend is also observed on DCTC In comparison the missingprobabilities of ADCT and HCTT keep zero for all the timeAs for ADCT the missing probability is not affected by the

0

10

20

30

40

50

60

70

80

90

100


Sens

ing

cove

rage

()

0 02 04 06 08 1

HCTTDPT

ADCTDCTC

Figure 11 The sensing coverage versus the prediction error for allalgorithms

prediction error if 119903119888 gt 2119903119904 since even in the case whenthe prediction error reaches 119903119904 all nodes in the monitoringregion of the target will be waked up HCTT does not relyon prediction algorithms to activate sensor nodes for targettracking Therefore the prediction error has no effect on theperformance of HCTT

613 Sensing Coverage As shown in Figure 11 the sensingcoverage of DPT and DCTC decreases when the predictionerror increases while the sensing coverage of HCTT andADCT is not affected It is due to the fact that increasing theprediction error results in a larger deviation of the dynamiccluster from the expected cluster which means a large anumber of nodes that should be waked up to sense thetarget are still staying in the sleep state We can also see thatthe sensing coverage of DPT performs the worst among allapproaches even when the prediction error is zeroThis is thereason why DPT has the worst missing probability amongall schemes Furthermore the sensing coverage of HCTT isnearly 100 which means that almost all nodes surroundingthe target contribute to the estimate of the targetrsquos locationTherefore the estimate of the targetrsquos location is in generalmore accurate and reliable than any of other three schemesNote that the trend of the sensing coverage in Figure 11 isopposite to the trend of the missing probability in Figure 10

614 Energy Consumption To evaluate the energy efficiencyof HCTT we compare the energy consumption of HCTT toother three approaches The energy consumption presentedhere does not include the energy consumed for the failurerecovery

As shown in Figure 12 the energy consumptions of DPTDCTC and ADCT behave some drops when the predictionerror increases Since increasing the prediction error resultsin the drop of the sensing coverage the energy consumption


20

30

40

50

60

70

80

HCTTDPT DCTC

02 04 06 08 10Prediction error (119903119904)

Ener

gy co

nsum

ptio

n (m

J)

ADCT

Figure 12 The Energy consumption versus the prediction error forall algorithms

drops accordingly as the number of nodes activated forsensing the target decreases DCTC consumes more energythan other schemes as it relies on a tree structure whichincurs more overhead than a one-hop cluster structure Incontrast DPT consumes the least energy among all schemesas it does not need to construct and dismiss dynamicclusters The energy consumptions of ADCT and HCTTstand between the ones of DPT and DCTC HCTT is not themost energy efficient scheme as there is a tradeoff betweenthe energy consumption and missing probabilitysensingcoverage Although DPT is the most energy efficient targettracking approach it suffers the boundary problem whichresults in a high probability of missing the target

Note that the performances between ADCT and HCTTare very close However HCTT is designed to solve theboundary problem in cluster-based networks which implic-itly takes the advantages of the cluster structure For exampledata can be easily routed from a node to the sink or anothernode with a low delay which is also important for targettracking Besides we do not consider the energy consump-tion of the failure recovery of these schemes when thenetwork loses the target Obviously the energy consumptionsof the failure recovery of DPT ADCT and DCTC are muchhigher than that of HCTT especially for that of DPT

62 In Multihop Static Clusters Scenario In this part wewill further evaluate the performance of HCTT in multihopcluster-based networks Sensor nodes can be organized intomulti-hop clusters via various approaches For example Linand Chu [29] proposed a multi-hop clustering techniqueusing hop distance to control the cluster structure instead ofthe number of nodes in a cluster A randomized distributedmulti-hop clustering algorithm for organizing the sensorsinto clusters is proposed in [30] In this paper we do not dis-cuss how to effectively construct multi-hop cluster structurebut just assume that sensor nodes are formed into multi-hop

0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters

100 seconds50 seconds

1 15 2 25 3 35 4Hop count of static clusters

Figure 13 The number of dynamic clusters versus the hop count ofstatic clusters in the HCTT algorithm

clusters Sensor nodes are organized as grid clusters with acluster head at the center Each node decides the hop countsto its cluster head according to its relative distance We usethe same configurations in multi-hop cluster-based networksas those in one-hop cluster-based networks except that thetransmission range of each node is fixed to be 40m and theprediction error is 04 119903119904

Since our proposed scheme is used to solve the boundaryproblem for cluster-based networks we only compare theperformance of HCTT to that of DPT in terms of hop count

621 Number of Dynamic Clusters Figure 13 shows theeffects of the hop count of each static cluster on the numberof dynamic clusters We can see that the number of generateddynamic clusters decreases as the hop count increases from 1

to 4This is due to the fact that it ismore frequent to encounterboundaries of static clusters if the sizes of clusters becomesmaller Besides we can see that the number of dynamicclusters increases as the movement duration of the targetincreases

622 Missing Probability Figure 14 shows the performanceof the missing probability versus the hop count of each staticcluster We can see that the missing probability of DPT dropsslightly when the hop count increases from 1 to 4 This isbecause that it ismore likely tomakewrong predictions aboutthe next working cluster when the sizes of static clusters aresmaller In comparison the missing probability of HCTTkeeps zero for all the time and does not be influenced by thehop count This validates that HCTT can solve the boundaryproblem not only for one-hop cluster-based networks butalso for multi-hop cluster-based networks

623 Sensing Coverage As shown in Figure 15 the sensingcoverage of DPT increases as the hop count increases from 1


0

10

20

30

40

50

60

70

80

90

100

Hop count of static clusters

HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()

Figure 14 The missing probability versus the hop count of staticclusters

0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()

Figure 15 The sensing coverage versus the hop count of staticclusters

to 4 This explains why the missing probability drops slightlyas the hop count increases The sensing coverage of HCTTkeeps nearly 100 for all the time which means almost allnodes sensing the target contribute to the estimate of thetargetrsquos location Consequently we can conclude that HCTTis robust to both the prediction error and the hop count ofstatic clusters

624 Energy Consumption Figure 16 shows the trend of theenergy consumptions of DPT and HCTT in terms of hopcount Both the energy consumptions of DPT and HCTTlargely increase as the hop count increases from 1 to 4 Thisis because that each node should send its sensing data via

0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)

Figure 16 The Energy consumption versus the hop count of staticclusters

a multi-hop path to its cluster head in the case of multi-hop static clusters An interesting thing is that althoughthe energy consumption of DPT is smaller than that ofHCTT in the case of one-hop clusters the simulation showsopposite results in the case of multi-hop clusters Moreoverthe difference between the energy consumptions of DPT andHCTT also increases as the hop count increases That isthe energy consumption of DPT increases faster than thatof HCTT The reason is that dynamic clusters generatedat the boundaries of clusters are just one-hop clusters Thecommunication overhead in such one-hop dynamic clustersis much less than that in multi-hop static clusters OverallHCTT becomes more energy efficient than DPT in the caseof multi-hop clusters

Summary From previous discussions we can conclude thatHCTT can solve the boundary problem not only for one-hop cluster-based networks but also for multi-hop cluster-based networks We can also conclude that when the sizeof clusters increases HCTT outperforms DPT in terms ofmissing probability sensing coverage and energy consump-tion Furthermore we find that HCTT is robust to boththe prediction error and the hop count of static clustersfor target tracking Therefore HCTT is an efficient mobilitymanagement approach for target tracking in cluster-basedsensor networks

7 Conclusions

Target tracking is a typical and important application ofWSNs However tracking a moving target in cluster-basedWSNs suffers the boundary problem when the target movesacross or along the boundaries of clusters In this paper wepropose a novel mobility management protocol for targettracking in cluster-based WSNs The protocol integrateson-demand dynamic clustering into scalable cluster-based


WSNs with the help of boundary nodes which facilitatessensorsrsquo collaboration among clusters and their smooth targettracking from one static cluster to another The proposedalgorithm solves the boundary problem of cluster-basedsensor networks and achieves a well tradeoff between energyconsumption and local node collaboration The simulationresults demonstrate the efficiency of the proposed protocol inboth one-hop and multi-hop cluster-based sensor networks

Acknowledgments

This work was supported in part by NSFC Grants no61273079 and no 61272463 Key Lab of Industrial ControlTechnology Grants no ICT1206 and no ICT1207 HongKongGRF Grants PolyU-524308 and PolyU-521312 HKPU GrantA-PL84 the Fundamental Research Funds for the CentralUniversities no 13CX02100A and Zhejiang Provincial KeyLab of Intelligent Processing Research of Visual Media no2012008

References

[1] I F Akyildiz W Su Y Sankarasubramaniam and E Cayirci ldquoAsurvey on sensor networksrdquo IEEE Communications Magazinevol 40 no 8 pp 102ndash105 2002

[2] C R Lin and M Gerla ldquoAdaptive clustering for mobilewireless sensor networksrdquo IEEE Journal on Selected Areas inCommunications vol 15 pp 1265ndash1275 1997

[3] M R Pearlman and Z J Haas ldquoDetermining the optimalconfiguration for the zone routing protocolrdquo IEEE Journal onSelected Areas in Communications vol 17 no 8 pp 1395ndash14141999

[4] U C Kozat G Kondylis B Ryu and M K Marina ldquoVirtualdynamic backbone for mobile ad hoc networksrdquo in Proceedingsof the International Conference on Communications (ICC rsquo01)pp 250ndash255 June 2000

[5] W P Chen J C Hou and L Sha ldquoDynamic clustering foracoustic target tracking in wireless sensor networksrdquo IEEETransactions on Mobile Computing vol 3 no 3 pp 258ndash2712004

[6] W C Yang Z Fu J H Kim and M S Park ldquoAn adaptivedynamic cluster-based protocol for target tracking in wirelesssensor networksrdquo in Advances in Data and Web Managementvol 4505 of Lecture Notes in Computer Science pp 156ndash167Springer Berlin Germany 2007

[7] W Zhang and G Cao ldquoDCTC dynamic convoy tree-basedcollaboration for target tracking in sensor networksrdquo IEEETransactions on Wireless Communications vol 3 no 5 pp1689ndash1701 2004

[8] X Ji H Zha J J Metzner and G Kesidis ldquoDynamic clusterstructure for object detection and tracking in wireless ad-hoc sensor networksrdquo in Proceedings of the IEEE InternationalConference on Communications pp 3807ndash3811 June 2004

[9] G Y Jin X Y Lu andM S Park ldquoDynamic clustering for objecttracking in wireless sensor networksrdquo in Proceedings of the 3rdIternational Conference on Ubiquitous Computing Systems (UCSrsquo06) pp 200ndash209 2006

[10] H Medeiros J Park and A C Kak ldquoDistributed objecttracking using a cluster-based Kalman filter in wireless cameranetworksrdquo IEEE Journal on Selected Topics in Signal Processingvol 2 no 4 pp 448ndash463 2008

[11] H Yang and B Sikdar ldquoA protocol for tracking mobile targetsusing sensor networksrdquo in Proceedings of the 1st IEEE Interna-tional Workshop on Sensor Network Protocols and Applicationspp 71ndash81 2003

[12] Z B Wang H B Li X F Shen X C Sun and Z WangldquoTracking and predicting moving targets in hierarchical sensornetworksrdquo in Proceedings of the IEEE International Conferenceon Networking Sensing and Control pp 1169ndash1174 2008

[13] Z B Wang Z Wang H L Chen J F Li and H B LildquoHiertrackmdashan energy efficient target tracking system for wire-less sensor networksrdquo in Proceedings of the 9th ACMConferenceon Embedded Networked Sensor Systems pp 377ndash378 2011

[14] W B Heinzelman A P Chandrakasan and H Balakrish-nan ldquoAn application-specific protocol architecture for wirelessmicrosensor networksrdquo IEEE Transactions onWireless Commu-nications vol 1 no 4 pp 660ndash670 2002

[15] Z BWangW Lou ZWang J C Ma andH L Chen ldquoA novelmobility management scheme for target tracking in cluster-based sensor networksrdquo in Proceedings of the IEEE InternationalConference onDistributedComputing in Sensor Systems pp 172ndash186 2010

[16] F Zhao J Shin and J Reich ldquoInformation-driven dynamicsensor collaboration for tracking applicationsrdquo IEEE SignalProcessing Magazine vol 19 no 2 pp 61ndash72 2002

[17] R R Brooks C Griffin and D S Friedlander ldquoSelf-organizeddistributed sensor network entity trackingrdquo International Jour-nal of High Performance Computing Applications vol 16 no 3pp 207ndash219 2002

[18] J Chen K Cao K Li and Y Sun ldquoDistributed sensor activationalgorithm for target tracking with binary sensor networksrdquoCluster Computing vol 14 no 1 pp 55ndash64 2011

[19] F Zhang J Chen H Li Y Sun and X M Shen ldquoDistributedactive sensor scheduling for target tracking in ultrasonic sensornetworksrdquo Mobile Networks and Applications vol 17 no 5 pp582ndash593 2011

[20] ZWang J A Luo and X P Zhang ldquoA novel location-penalizedmaximum likelihood estimator for bearing-only target localiza-tionrdquo IEEE Transaction on Signal Processing vol 60 no 12 pp6166ndash6181 2012

[21] T He S Krishnamurthy L Luo et al ldquoVigilNet an integratedsensor network system for energy-efficient surveillancerdquo ACMTransactions on Sensor Networks vol 2 no 1 pp 1ndash38 2006

[22] T He P Vicaire T Yant et al ldquoAchieving real-time targettracking using wireless sensor networksrdquo in Proceedings of the12th IEEEReal-Time andEmbeddedTechnology andApplicationsSymposium pp 37ndash48 April 2006

[23] P Cheng J M Chen F Zhang Y X Sun and X M ShenldquoA distributed TDMA scheduling algorithm for target trackingin ultrasonic sensor networksrdquo IEEE Transactions on IndustrialElectronics no 99 2012

[24] H Chen W Lou and Z Wang ldquoA novel secure localizationapproach in wireless sensor networksrdquo Eurasip Journal onWireless Communications and Networking vol 2010 Article ID981280 12 pages 2010

[25] M Fayyaz ldquoClassification of object tracking techniques inwireless sensor networksrdquo Wireless Sensor Network vol 3 pp121ndash124 2011

[26] O Demigha W-K Hidouci and T Ahmed ldquoOn energyefficiency in collaborative target tracking in wireless sensornetwork a reviewrdquo IEEE Communications Surveys amp Tutorialsvol 99 pp 1ndash13 2012


[27] O Younis and S Fahmy ldquoHEED a hybrid energy-efficientdistributed clustering approach for ad hoc sensor networksrdquoIEEE Transactions on Mobile Computing vol 3 no 4 pp 366ndash379 2004

[28] J Y Yu and P H J Chong ldquoA survey of clustering schemesfor mobile ad hoc networksrdquo IEEE Communications Surveys ampTutorials vol 7 no 1 pp 32ndash48 2005

[29] H C Lin and Y H Chu ldquoClustering technique for largemultihop mobile wireless networksrdquo in Proceedings of the 51stVehicular Technology Conference (VTC rsquo00) pp 1545ndash1549 May2000

[30] A Youssef M Younis M Youssef and A Agrawala ldquoDis-tributed formation of overlappingmulti-hop clusters in wirelesssensor networksrdquo in Proceedings of the IEEE Global Telecom-munications Conference Exhibition amp Industry Forum (GLOBE-COM rsquo06) December 2006

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



head so as to balance the energy load) It uses a predictivemechanism to inform cluster heads about the approachingtarget and then the corresponding cluster head wakes upa number of appropriate nodes right before the arrival ofthe target The overhead can be saved as the tracking taskis handed over from one static cluster to another with-out costly dynamic clustering processes It also provides ascalable structure for coordinating and managing networksTherefore static cluster-based approaches are more suitablefor target tracking in large-scale sensor networks Howeverthe static cluster membership prevents sensors in differentclusters from collaborating and sharing information witheach other which causes a so-called boundary problem whenthe target moves across or along the boundaries of clustersThe boundary problem will result in the increase of trackinguncertainty or even the loss of the target Therefore a newprotocol is required to solve the boundary problem and realizethe tradeoff between energy consumption and local sensorcollaboration for cluster-based sensor networks

In this paper we propose a novel distributed mobil-ity management protocol called hybrid cluster-based targettracking (HCTT) for efficient target tracking in a large-scalecluster-based WSN HCTT integrates on-demand dynamicclustering into a scalable cluster-based WSN with the helpof boundary nodes which facilitates sensorsrsquo collaborationamong clusters to solve the boundary problem As shown inFigure 1 when the target is inside a static cluster the clusteris responsible for target tracking as the target moves close tothe boundaries of clusters an on-demand dynamic clusteringprocess will be triggered to manage the tracking task so asto avoid the boundary problem The on-demand dynamiccluster will dismiss soon after the target moves away fromthe boundaries As shown in Figure 1 the sequential clustersfor tracking the target are 119860 rarr 1198631 rarr 119862 rarr 1198632 rarr

119864 When the target moves static clusters and on-demanddynamic clusters alternately manage the tracking task Byintegrating on-demand dynamic clustering into a scalablecluster-based structure nodes belonging to different staticclusters can share information which guarantees smoothlytarget tracking and realizes well tradeoff between energyconsumption and local sensor collaboration Note that aconference paper [15] containing some preliminary resultsof this paper appeared at IEEE DCOSS 2010 This paper hasbeen revised by following the comments of the reviewersand the feedback of other researchers Moreover we analyzethe computation and communication complexity of HCTTand extend the performance evaluation to multihop clusterscenario as opposed to just one-hop scenario

The main contributions of this paper are summarized asfollows

(i) We address the boundary problem for target trackingin a cluster-based WSN

(ii) We present a novel hybrid cluster-based target track-ing protocol which balances well between the energyconsumption and local sensor collaboration

(iii) The proposed algorithm is scalable to multi-hopstatic clusters and solves the problem of trackinguncertainty due to prediction errors

Dynamic cluster

Static cluster

Cluster headMember nodeLeader of dynamic cluster

119861

119860

119862

1198631

1198632

119864

119865

119863

Figure 1 Illustration of HCTT for target tracking in a cluster-basedWSN

(iv) We conduct simulations to show the efficiency of theproposed scheme compared with other typical targettracking solutions

The rest of the paper is organized as follows Section 2reviews the related work about target tracking in WSNs Sec-tion 3 describes the systemmodel and the boundary problemfor target tracking in a cluster-based WSN In Section 4we present the hybrid cluster-based target tracking protocolin detail The analysis of HCTT is presented in Section 5Section 6 demonstrates the performance evaluation for one-hop clusters and multi-hop clusters Finally we conclude thepaper in Section 7

2 Related Work

Target tracking in WSNs has received considerable atten-tion from various angles and a lot of protocols havebeen proposed Zhao et al proposed an information-drivendynamic sensor collaboration mechanism for target tracking[16] Brooks et al presented a distributed entity-trackingframework for sensor networks [17] Chen et al proposeddistributed sensor scheduling algorithms for binary sensornetworks [18] and acoustic sensor networks [19] respectivelyWang et al proposed a novel localization algorithm fortarget tracking [20] Vigilnet an energy-efficient and real-time integrated system for target tracking was designed andimplemented in [21 22] A distributed TDMA schedulingalgorithm for target tracking in ultrasonic sensor networkswas proposed in [23] In [24] the authors proposed a securelocalization scheme to defend against some typical attacksthat may affect target tracking accuracy More target trackingprotocols can be found in [25 26]

To balance energy consumption and sensor collaborationa lot of dynamic clustering protocols have been proposedfor target tracking in WSNs Yang et al proposed anadaptive dynamic cluster-based tracking (ADCT) protocol








119875119903 (119873 (119860) = 119896) =(120582119860)119896

119896sdot (119890minus120582119860

) 119896 = 0 1 infin (1)



119903119894 =

120573

119889120572 (119897119894 119871) 119889 (119897119894 119871) le 119903119904

0 (119897119894 119871) gt 119903119904

(2)









Monitoring region

119861

119889

119888

119887

119886

119891119890

119860

119862

CH

CH

CH

(a)

Monitoring region

Actual location

Predicted location

119861

119889

119888

119887

119886

119891119890

119860

119862

CH

CH

CH

(b)







119903119888 and 119894 = 119895


119889(119897119894 119897119895) le 119903119888


119894=1119862119894 = 119881 and 119862119894⋂119862119895 = 120601 where 119894 = 119895 | 119894 119895 =

1 2 119898


V1198671

V1198672

V119867119898

sub 119881 where V119867119894











Dynamicclustering

D2SICHandoff

S2DIC

HandoffD2DIC

Handoff

HCTT










and 119862(V119894) = 119862(V119895)




119877119878 (119862119894) = ⋃

forallV119894isin119868(119862119894)

119877 (V119894 119903119904) minus ⋃

forallV119894isin119861(119862119894)

119877 (V119894 119903119904) (3)


119877119861 (119862119894) = ⋃


(119877 (V119894 119903119904) cap 119877 (V119895 119903119904))

(4)




119894) 119894 = 1 2 119898



119862(V119895) = 119862(V119896) do(5) V119895 sdot 119904119905119886119905119890 larr 119887119900119906119899119889119886119903119910

(6) 119861 (119862119894) larr 119861 (119862119894) cup V119895


Alert region

Boundary region


Safety region



119877119860 (119862119894) = ⋃

forallV119894isin119861(119862119894)

119877 (V119894 119903119904) minus 119877119861 (119862119894) (5)













1198631


119860

119861

119862119863

119886

119887119888

119889

119890

(a)


119860

119861

119862

119863

1198633

1198632119898

119899

119902

119901119900

(b)






























119860

119886

119888

119887

11987931198792

1198791

1198634119861












5 Analysis of HCTT



119888minus 1


Request

Handoff

AcknowledgeWork

Sleep

CH119894 119871dc

(a)

Handoff

AcknowledgeWork

Dismiss

CH119895119871dc

(b)

Request

Handoff

AcknowledgeWork

Dismiss

119871dc998400119871dc

(c)




119888minus 1










2

119888minus




2














2

119888 Typically








0 100 200 300 400

50

100

150

200

250

300

350







2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

Num

ber o

f dyn

amic

clus

ters

119903119888119903119904


00

10

20

30

40

50

60

70

80

90

100

HCTTDPT

ADCTDCTC

02 04 06 08 1

Miss

pro

babi

lity

()





0

10

20

30

40

50

60

70

80

90

100


Sens

ing

cove

rage

()

0 02 04 06 08 1

HCTTDPT

ADCTDCTC







20

30

40

50

60

70

80

HCTTDPT DCTC


Ener

gy co

nsum

ptio

n (m

J)

ADCT





0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters











0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation
















119875119903 (119873 (119860) = 119896) =(120582119860)119896

119896sdot (119890minus120582119860

) 119896 = 0 1 infin (1)



119903119894 =

120573

119889120572 (119897119894 119871) 119889 (119897119894 119871) le 119903119904

0 (119897119894 119871) gt 119903119904

(2)









Monitoring region

119861

119889

119888

119887

119886

119891119890

119860

119862

CH

CH

CH

(a)

Monitoring region

Actual location

Predicted location

119861

119889

119888

119887

119886

119891119890

119860

119862

CH

CH

CH

(b)







119903119888 and 119894 = 119895


119889(119897119894 119897119895) le 119903119888


119894=1119862119894 = 119881 and 119862119894⋂119862119895 = 120601 where 119894 = 119895 | 119894 119895 =

1 2 119898


V1198671

V1198672

V119867119898

sub 119881 where V119867119894











Dynamicclustering

D2SICHandoff

S2DIC

HandoffD2DIC

Handoff

HCTT










and 119862(V119894) = 119862(V119895)




119877119878 (119862119894) = ⋃

forallV119894isin119868(119862119894)

119877 (V119894 119903119904) minus ⋃

forallV119894isin119861(119862119894)

119877 (V119894 119903119904) (3)


119877119861 (119862119894) = ⋃


(119877 (V119894 119903119904) cap 119877 (V119895 119903119904))

(4)




119894) 119894 = 1 2 119898



119862(V119895) = 119862(V119896) do(5) V119895 sdot 119904119905119886119905119890 larr 119887119900119906119899119889119886119903119910

(6) 119861 (119862119894) larr 119861 (119862119894) cup V119895


Alert region

Boundary region


Safety region



119877119860 (119862119894) = ⋃

forallV119894isin119861(119862119894)

119877 (V119894 119903119904) minus 119877119861 (119862119894) (5)













1198631


119860

119861

119862119863

119886

119887119888

119889

119890

(a)


119860

119861

119862

119863

1198633

1198632119898

119899

119902

119901119900

(b)






























119860

119886

119888

119887

11987931198792

1198791

1198634119861












5 Analysis of HCTT



119888minus 1


Request

Handoff

AcknowledgeWork

Sleep

CH119894 119871dc

(a)

Handoff

AcknowledgeWork

Dismiss

CH119895119871dc

(b)

Request

Handoff

AcknowledgeWork

Dismiss

119871dc998400119871dc

(c)




119888minus 1










2

119888minus




2














2

119888 Typically








0 100 200 300 400

50

100

150

200

250

300

350







2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

Num

ber o

f dyn

amic

clus

ters

119903119888119903119904


00

10

20

30

40

50

60

70

80

90

100

HCTTDPT

ADCTDCTC

02 04 06 08 1

Miss

pro

babi

lity

()





0

10

20

30

40

50

60

70

80

90

100


Sens

ing

cove

rage

()

0 02 04 06 08 1

HCTTDPT

ADCTDCTC







20

30

40

50

60

70

80

HCTTDPT DCTC


Ener

gy co

nsum

ptio

n (m

J)

ADCT





0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters











0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Monitoring region

119861

119889

119888

119887

119886

119891119890

119860

119862

CH

CH

CH

(a)

Monitoring region

Actual location

Predicted location

119861

119889

119888

119887

119886

119891119890

119860

119862

CH

CH

CH

(b)







119903119888 and 119894 = 119895


119889(119897119894 119897119895) le 119903119888


119894=1119862119894 = 119881 and 119862119894⋂119862119895 = 120601 where 119894 = 119895 | 119894 119895 =

1 2 119898


V1198671

V1198672

V119867119898

sub 119881 where V119867119894











Dynamicclustering

D2SICHandoff

S2DIC

HandoffD2DIC

Handoff

HCTT










and 119862(V119894) = 119862(V119895)




119877119878 (119862119894) = ⋃

forallV119894isin119868(119862119894)

119877 (V119894 119903119904) minus ⋃

forallV119894isin119861(119862119894)

119877 (V119894 119903119904) (3)


119877119861 (119862119894) = ⋃


(119877 (V119894 119903119904) cap 119877 (V119895 119903119904))

(4)




119894) 119894 = 1 2 119898



119862(V119895) = 119862(V119896) do(5) V119895 sdot 119904119905119886119905119890 larr 119887119900119906119899119889119886119903119910

(6) 119861 (119862119894) larr 119861 (119862119894) cup V119895


Alert region

Boundary region


Safety region



119877119860 (119862119894) = ⋃

forallV119894isin119861(119862119894)

119877 (V119894 119903119904) minus 119877119861 (119862119894) (5)













1198631


119860

119861

119862119863

119886

119887119888

119889

119890

(a)


119860

119861

119862

119863

1198633

1198632119898

119899

119902

119901119900

(b)






























119860

119886

119888

119887

11987931198792

1198791

1198634119861












5 Analysis of HCTT



119888minus 1


Request

Handoff

AcknowledgeWork

Sleep

CH119894 119871dc

(a)

Handoff

AcknowledgeWork

Dismiss

CH119895119871dc

(b)

Request

Handoff

AcknowledgeWork

Dismiss

119871dc998400119871dc

(c)




119888minus 1










2

119888minus




2














2

119888 Typically








0 100 200 300 400

50

100

150

200

250

300

350







2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

Num

ber o

f dyn

amic

clus

ters

119903119888119903119904


00

10

20

30

40

50

60

70

80

90

100

HCTTDPT

ADCTDCTC

02 04 06 08 1

Miss

pro

babi

lity

()





0

10

20

30

40

50

60

70

80

90

100


Sens

ing

cove

rage

()

0 02 04 06 08 1

HCTTDPT

ADCTDCTC







20

30

40

50

60

70

80

HCTTDPT DCTC


Ener

gy co

nsum

ptio

n (m

J)

ADCT





0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters











0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation














Dynamicclustering

D2SICHandoff

S2DIC

HandoffD2DIC

Handoff

HCTT










and 119862(V119894) = 119862(V119895)




119877119878 (119862119894) = ⋃

forallV119894isin119868(119862119894)

119877 (V119894 119903119904) minus ⋃

forallV119894isin119861(119862119894)

119877 (V119894 119903119904) (3)


119877119861 (119862119894) = ⋃


(119877 (V119894 119903119904) cap 119877 (V119895 119903119904))

(4)




119894) 119894 = 1 2 119898



119862(V119895) = 119862(V119896) do(5) V119895 sdot 119904119905119886119905119890 larr 119887119900119906119899119889119886119903119910

(6) 119861 (119862119894) larr 119861 (119862119894) cup V119895


Alert region

Boundary region


Safety region



119877119860 (119862119894) = ⋃

forallV119894isin119861(119862119894)

119877 (V119894 119903119904) minus 119877119861 (119862119894) (5)













1198631


119860

119861

119862119863

119886

119887119888

119889

119890

(a)


119860

119861

119862

119863

1198633

1198632119898

119899

119902

119901119900

(b)






























119860

119886

119888

119887

11987931198792

1198791

1198634119861












5 Analysis of HCTT



119888minus 1


Request

Handoff

AcknowledgeWork

Sleep

CH119894 119871dc

(a)

Handoff

AcknowledgeWork

Dismiss

CH119895119871dc

(b)

Request

Handoff

AcknowledgeWork

Dismiss

119871dc998400119871dc

(c)




119888minus 1










2

119888minus




2














2

119888 Typically








0 100 200 300 400

50

100

150

200

250

300

350







2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

Num

ber o

f dyn

amic

clus

ters

119903119888119903119904


00

10

20

30

40

50

60

70

80

90

100

HCTTDPT

ADCTDCTC

02 04 06 08 1

Miss

pro

babi

lity

()





0

10

20

30

40

50

60

70

80

90

100


Sens

ing

cove

rage

()

0 02 04 06 08 1

HCTTDPT

ADCTDCTC







20

30

40

50

60

70

80

HCTTDPT DCTC


Ener

gy co

nsum

ptio

n (m

J)

ADCT





0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters











0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











119894) 119894 = 1 2 119898



119862(V119895) = 119862(V119896) do(5) V119895 sdot 119904119905119886119905119890 larr 119887119900119906119899119889119886119903119910

(6) 119861 (119862119894) larr 119861 (119862119894) cup V119895


Alert region

Boundary region


Safety region



119877119860 (119862119894) = ⋃

forallV119894isin119861(119862119894)

119877 (V119894 119903119904) minus 119877119861 (119862119894) (5)













1198631


119860

119861

119862119863

119886

119887119888

119889

119890

(a)


119860

119861

119862

119863

1198633

1198632119898

119899

119902

119901119900

(b)






























119860

119886

119888

119887

11987931198792

1198791

1198634119861












5 Analysis of HCTT



119888minus 1


Request

Handoff

AcknowledgeWork

Sleep

CH119894 119871dc

(a)

Handoff

AcknowledgeWork

Dismiss

CH119895119871dc

(b)

Request

Handoff

AcknowledgeWork

Dismiss

119871dc998400119871dc

(c)




119888minus 1










2

119888minus




2














2

119888 Typically








0 100 200 300 400

50

100

150

200

250

300

350







2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

Num

ber o

f dyn

amic

clus

ters

119903119888119903119904


00

10

20

30

40

50

60

70

80

90

100

HCTTDPT

ADCTDCTC

02 04 06 08 1

Miss

pro

babi

lity

()





0

10

20

30

40

50

60

70

80

90

100


Sens

ing

cove

rage

()

0 02 04 06 08 1

HCTTDPT

ADCTDCTC







20

30

40

50

60

70

80

HCTTDPT DCTC


Ener

gy co

nsum

ptio

n (m

J)

ADCT





0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters











0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










1198631


119860

119861

119862119863

119886

119887119888

119889

119890

(a)


119860

119861

119862

119863

1198633

1198632119898

119899

119902

119901119900

(b)






























119860

119886

119888

119887

11987931198792

1198791

1198634119861












5 Analysis of HCTT



119888minus 1


Request

Handoff

AcknowledgeWork

Sleep

CH119894 119871dc

(a)

Handoff

AcknowledgeWork

Dismiss

CH119895119871dc

(b)

Request

Handoff

AcknowledgeWork

Dismiss

119871dc998400119871dc

(c)




119888minus 1










2

119888minus




2














2

119888 Typically








0 100 200 300 400

50

100

150

200

250

300

350







2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

Num

ber o

f dyn

amic

clus

ters

119903119888119903119904


00

10

20

30

40

50

60

70

80

90

100

HCTTDPT

ADCTDCTC

02 04 06 08 1

Miss

pro

babi

lity

()





0

10

20

30

40

50

60

70

80

90

100


Sens

ing

cove

rage

()

0 02 04 06 08 1

HCTTDPT

ADCTDCTC







20

30

40

50

60

70

80

HCTTDPT DCTC


Ener

gy co

nsum

ptio

n (m

J)

ADCT





0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters











0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation





























119860

119886

119888

119887

11987931198792

1198791

1198634119861












5 Analysis of HCTT



119888minus 1


Request

Handoff

AcknowledgeWork

Sleep

CH119894 119871dc

(a)

Handoff

AcknowledgeWork

Dismiss

CH119895119871dc

(b)

Request

Handoff

AcknowledgeWork

Dismiss

119871dc998400119871dc

(c)




119888minus 1










2

119888minus




2














2

119888 Typically








0 100 200 300 400

50

100

150

200

250

300

350







2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

Num

ber o

f dyn

amic

clus

ters

119903119888119903119904


00

10

20

30

40

50

60

70

80

90

100

HCTTDPT

ADCTDCTC

02 04 06 08 1

Miss

pro

babi

lity

()





0

10

20

30

40

50

60

70

80

90

100


Sens

ing

cove

rage

()

0 02 04 06 08 1

HCTTDPT

ADCTDCTC







20

30

40

50

60

70

80

HCTTDPT DCTC


Ener

gy co

nsum

ptio

n (m

J)

ADCT





0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters











0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Request

Handoff

AcknowledgeWork

Sleep

CH119894 119871dc

(a)

Handoff

AcknowledgeWork

Dismiss

CH119895119871dc

(b)

Request

Handoff

AcknowledgeWork

Dismiss

119871dc998400119871dc

(c)




119888minus 1










2

119888minus




2














2

119888 Typically








0 100 200 300 400

50

100

150

200

250

300

350







2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

Num

ber o

f dyn

amic

clus

ters

119903119888119903119904


00

10

20

30

40

50

60

70

80

90

100

HCTTDPT

ADCTDCTC

02 04 06 08 1

Miss

pro

babi

lity

()





0

10

20

30

40

50

60

70

80

90

100


Sens

ing

cove

rage

()

0 02 04 06 08 1

HCTTDPT

ADCTDCTC







20

30

40

50

60

70

80

HCTTDPT DCTC


Ener

gy co

nsum

ptio

n (m

J)

ADCT





0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters











0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation















2

119888 Typically








0 100 200 300 400

50

100

150

200

250

300

350







2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

Num

ber o

f dyn

amic

clus

ters

119903119888119903119904


00

10

20

30

40

50

60

70

80

90

100

HCTTDPT

ADCTDCTC

02 04 06 08 1

Miss

pro

babi

lity

()





0

10

20

30

40

50

60

70

80

90

100


Sens

ing

cove

rage

()

0 02 04 06 08 1

HCTTDPT

ADCTDCTC







20

30

40

50

60

70

80

HCTTDPT DCTC


Ener

gy co

nsum

ptio

n (m

J)

ADCT





0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters











0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

Num

ber o

f dyn

amic

clus

ters

119903119888119903119904


00

10

20

30

40

50

60

70

80

90

100

HCTTDPT

ADCTDCTC

02 04 06 08 1

Miss

pro

babi

lity

()





0

10

20

30

40

50

60

70

80

90

100


Sens

ing

cove

rage

()

0 02 04 06 08 1

HCTTDPT

ADCTDCTC







20

30

40

50

60

70

80

HCTTDPT DCTC


Ener

gy co

nsum

ptio

n (m

J)

ADCT





0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters











0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










20

30

40

50

60

70

80

HCTTDPT DCTC


Ener

gy co

nsum

ptio

n (m

J)

ADCT





0

5

10

15

20

25

30

35

40

45

50

Num

ber o

f dyn

amic

clus

ters











0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Miss

pro

babi

lity

()


0

10

20

30

40

50

60

70

80

90

100


HCTTDPT

1 15 2 25 3 35 4

Sens

ing

cove

rage

()




0

20

40

60

80

100

120

140

160

180

200


HCTTDPT

1 15 2 25 3 35 4

Ener

gy co

nsum

ptio

n (m

J)




7 Conclusions




Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











Acknowledgments


References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation
















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Research Article A Hybrid Cluster-Based Target Tracking...

Documents

Transcript of Research Article A Hybrid Cluster-Based Target Tracking...