RSSI-Based Smooth Localization for Indoor Environment

Research ArticleRSSI-Based Smooth Localization for Indoor Environment

Yujian Wang Bin Zhao and Zhaohui Jiang

IPPFRD Alcatel-Lucent Shanghai Bell Co Ltd Shanghai 201206 China

Correspondence should be addressed to Yujian Wang yujianwangalcatel-sbellcomcn

Received 9 May 2014 Accepted 3 June 2014 Published 29 June 2014

Academic Editor Hongli Xu

Copyright copy 2014 Yujian 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

Radio frequency (RF) technique for its better penetrability over traditional techniques such as infrared or ultrasound is widelyused for indoor localization and tracking In this paper three novel measurements point decision accuracy path matching errorand wrong jumping ratio are firstly defined to express the localization efficiency Then a novel RSSI-based smooth localization(RSL) algorithm is designed implemented and evaluated on theWiFi networksThe tree-basedmechanism determines the currentposition and track of the entity by assigning the weights and accumulative weights for all collected RSSI information of referencepoints so as to make the localization smoothThe evaluation results indicate that the proposed algorithm brings better localizationsmoothness of reducing 10 path matching error and 30 wrong jumping ratio over the RADAR system

1 Introduction

Nowadays location is the important information for manyinnovative applications [1] For example smart asset manage-ment (SAM) system requires the position of each asset Asthe asset is moved from one place to another the managershould master the physical movement in a real-time mannerWith location information the system helps to reduce thesearching time of each asset and improve its usage efficiencyfor the enterprise Obviously it is of significant importance todetermine the real-time position of each mobile entity (egan asset) which is called indoor localization or tracking

The techniques for indoor tracking are classified into twobroad categories non-RF and RF techniques In the formercategory non-RF methods such as infrared or ultrasoundmay be used alone or together with RF for indoor tracking [23] From the view of marketing the additional equipmentssuch as accelerometer compass and gyroscope increase thecost of the large-scale deployment Besides these techniquesmay notworkwell in the complicated indoor environment forits worse penetration For instance the signals will be blockedwhen the devices are buried in the usersrsquo wallets or bags [1]

As RF measurement (eg RSSI) will be easily obtainedduring the wireless communication there are many RSSI-based methods for indoor localization Previous work mostlycares about the position error which denotes the averagedistance between the real position and located position

However this measurement cannot fully reflect the perfor-mance for continuous localization in practice For examplethe entity is moving from point 119860 to point 119861 Whereas asthe instability and unreliability of RSSI the obtained trackingpath is 119860-119862-119861 though point 119862 does not lie between point 119860and point 119861 In this case though the one-shot wrong jumpingmay not greatly increase the average position error this willresult in wrong path determination So that the users mayregard the localization result as unreasonable for the wrongtracking path To express the tracking efficiency we call itsmooth localization if the computed path by the algorithmis exactly the same as the real path for a mobile entityHowever the multipath reflection and wireless interferencebring some challenges to implement the smooth localization(1) RSSI is not stable The value of RSSI is determined bythe transmission power distance transmission path and soforth Even between two fixed nodes the RSSI value variesas time goes by (2) RSSI is not always reliable Transmissionreliability will change under different scenarios When thenumber of mobile entities is increasing the communica-tion interference increases which reduces the transmissionreliability (3) There is no causal relationship between RSSIand Euclidean distance so that large distance estimation isbrought about by judging from RSSI information only

The researchers have designed many algorithms to con-quer one or some of the above challenges for indoor localiza-tion For example as there is no apparent function between

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 639142 8 pageshttpdxdoiorg1011552014639142

2 The Scientific World Journal

RSSI and Euclidean distance several works [4] compute thenodesrsquo localizations based on the connectivity informationThen RSSI is used to enhance the location precision [5 6]However there is mere work dealing with the unreliabil-ity and instability of RSSI which make many algorithmsunfit for the practical applications The existing RF-basedtracking systems mostly focus on the precise position oraverage position error However no apparent performancemeasurement has been studied for smooth localization Forexample the MERIT system [7] defines a new measurementfor indoor localization called area decision accuracy todetermine where the entity locates by comparing the averageRSSIs of different areas As RSSI is not always stable orreliable it does not work well under the indoor environmentMERIT [7] requires the denser AP deployment and doesnot fit for the areas with different area sizes As in the SAMapplication wrong area decision will result in wrong assetsearching and asset statistics With the negative features ofRSSI smooth localization becomes an extremely difficultchallenge for researchers As we know there is no specialwork aiming to smooth localization The main contributionof this paper is as follows

(i) Three novel measurements point decision accuracypath matching error and wrong jumping ratio aredefined to express the localization smoothness forindoor environment

(ii) A RSSI-based smooth localization (RSL) algorithm isproposed for the indoor localization The algorithmfirst selects some reference points and constructsa connected graph for these points based on thearchitectural features Then reference points will beassigned weights and accumulative weights by thecollectedRSSI information Finally a novel tree-basedmechanism is designed to estimate the moving trackwith high probability for smooth localization

(iii) The proposed algorithm is validated on the WiFi testbedThe evaluation results indicate that the proposedalgorithm can get better localization smoothnesscompared with the previous works For example theRSL algorithm can reduce the path matching errorand wrong jumping ratio by about 10 and 30 overthe RADAR system

In Section 2 the localization framework is introducedand three novel performance measurements for smoothlocalization are defined The RSSI-based smooth localizationalgorithm is discussed in detail in Section 3The validation isdescribed in Section 4 according to the above measurementsFinally the conclusion is drawn in Section 5

2 Localization Performance Measurement

21 Localization Framework Many algorithms of previousworkmay locate themobile entity to arbitrary positions in thefield However as RSSI information is unstable and wirelesstransmission apt to be unreliable the traditional RSSI-basedmethods might result in unstable localization That is thelocalization result will jump from one point to another while

AB

u1

u2

u3

Figure 1 Illustration of localization measurement

the entity is stationary to a fixed spot To provide smoothlocalization this procedure mainly consists of two steps(1) RSSI sampling and (2) position determination Like theprevious method [8] the localization framework first selectsmany reference points (also called representational points)in the field For example the central point of a room will beselected to represent this roomThe system samples the RSSIinformation on these discrete points for a durative periodIn the runtime the mobile entity will be located on somepresampled positions by the collected RSSI informationIn the second step different mechanisms are designed toimprove the localization performanceThemain advantage ofthis framework is to increase the localization efficiency andstability compared with the previous mechanisms [7] Thetask of this paper is to design efficient methods to improvethe localization smoothness under this practical framework

22 LocalizationMeasurement Definition Theprevious algo-rithms [5 7ndash9] mostly take the average position errorwhich denotes the distance between the real position andlocated position as the main performance measurementHowever this measurement cannot often reflect the realeffect for indoor localization For example there are tworooms in a field shown in Figure 1 Assume that the currentposition of a mobile entity is 119906

1 Two localization algorithms

obtain the different position results denoted by and throughRSSI information From the view of average position errorposition 119906

2is closer to 119906

1than to 119906

3 so that theoretically

the average position error between 1199061and 119906

2is smaller than

that of 1199061and 119906

3 However 119906

1is located in the same room as

1199063 Thus it is more reasonable to locate the mobile entity on

position 1199063compared to position 119906

2

In the following we will define some practical perfor-mance measurements for indoor localization There are twodifferent scenes for indoor localization One is static theother is mobile

For static scenes the entitymostly stays at one position oraround for a long time Under the above localization frame-work each position in the field will be logically appointed to areference point For example two points are in the same areaor close to each other We adopt the point decision accuracy(PDA) to measure the localization efficiency under the staticscenes In this case point decision accuracy is defined as thecorrect probability that the mobile entity is located at theappointed reference point by the localization algorithm Forexample as shown in Figure 1 we regard that each point in

The Scientific World Journal 3

the area119860 is appointed to position 1199063 for they lie in the same

room The system collects the RSSI information for 100 timeslots as the mobile entity is on point 119906

1 If there are 90 time

slots in which the localization result is 1199063 we regard that the

point decision accuracy of this algorithm is about 09 If eacharea only contains one reference point point decision is thesame as area decision [7] Thus PDA is more general thanarea decision accuracy

For mobile scenes the entity continuously moves fromone position to another However one may stay on a certainpoint for a short time compared with a long period Toreflect the effect of the practical movement we will definetwo measurements to express the localization smoothnessOne is called as path matching error (PME) We considerthe discrete-time division Assume that entity 119906 locates ona point 119901(119905) in time slot 119905 Note that the mobile entitymay not locate on a reference point In comparison as eachposition will be logically appointed to a reference point inthe field we use the appointed reference point to denote thereal position That is 119901(119905) is a reference point to which thereal position of entity 119906 may be appointed Then a real pathcan be described as a position string RP = 119901(1) 119901(119899)where 119899 is the total number of time slots After informationprocessing the server may obtain the localization result119903(119905) in time slot 119905 Note that 119903(119905) is a reference pointunder the above localization framework Thus a localizationresult path denoted by LP = 119903(1) 119903(119899) is obtainedTo express the similarity between two paths we computetheir maximum common substring of two paths RP and LPdenoted by MCS(RP LP) Moreover its length is denoted by|MCS(RP LP)| Then PME is calculated as

PME (RP LP) = 1 minus|MCS (RP LP)|

119899 (1)

The other measurement is called wrong jumping ratio(WJR) The unstable and unreliable RSSI information willresult in wrong jumping which greatly debases the localiza-tion Assume that the real track of a mobile entity is ldquo1133rdquoWith RSSI information the localization result is ldquo1123rdquo Sothere is one wrong jumping from point 1 to point 2 Note thatthis jumping will result in wrong path computation WJR isdefined as the average number of wrong localization jumpingduring 100 time slots For example the system collects theRSSI information and computes the localization for 119899 timeslots The number of wrong localization jumping is 119898 ThenWJR is defined as

WJR =119898

119899times 100 (2)

The smooth localization aims to reduce the pathmatchingerror and to avoid the wrong localization jumping Now wegive examples to illustrate the definitions of PME and WJRFor example the real path consists of some discrete positionsand is expressed by RP = ldquo1112233345rdquo Assume that the firstalgorithm derives the path as LP

1= ldquo1112223345rdquo By the

definition the maximum common substring of RP and LP1

is ldquo111223345rdquo and its length is 9 So its path matching erroris 01 Though LP

1cannot fully be matched to RP there is

Figure 2 Access point

no wrong jumping between two paths Thus the WJR of LP1

is 0 This algorithm just locates the entity to point 3 witha delay Assume the second algorithm obtains a path LP

2=

ldquo1142223345rdquo Similarly the maximum common substring ofRP and LP

2is ldquo11223345rdquo and its length is 8 As a result its

path matching error is 02 Observing the localization resultLP2 there is one wrong localization jumping from point 1 to

point 4 and itsWJR is (110)times100 = 10The third algorithmobtains a path LP

3= ldquo1142233245rdquo The maximum common

substring of RP andLP3is ldquo11223345rdquo and its length is 8Then

its path matching error is 02 However there are two wrongjumping from point 1 to point 4 and from point 3 to point 2respectively Its WJR is (210) times 100 = 20 According to theabove description we regard that LP

1is much smoother than

paths LP2and LP

3Though LP

2and LP

3reach the same PME

LP2will work better than LP

3for WJR

3 Smooth Localization Algorithm Description

In this section we will design a RSSI-based smooth local-ization (RSL) algorithm which aims to obtain higher pointdecision accuracy lower path matching error and lowerwrong jumping ratio The RSL algorithm consists of threesteps reference point selection graph construction andlocalization determination respectively

31 System Equipments Before algorithm description wefirst introduce the main equipment in the system

There are two main categories of equipments in thesystem One is access point (AP) shown in Figure 2 Each APcan support the standard 80211 bgn and is used to collectthe detection packets from all themobile entities in each timeslot Moreover all access points can form a wireless backbonenetwork and transmit the obtained RSSI information to thelocalization server As WiFi networks are widely deployedthe localization system can be built on these wireless net-works which helps to save deployment costs The otheris WiFi-compatible card shown in Figure 3 Each mobileentity will take a card which locally broadcasts the detectionpackets to the neighboring access points In the system eachcard is only capable of wireless transmission which is energyefficient cheap and fit for large-scale applications After RSSI


Figure 3 WiFi-compatible card

12

10

11

9

8

7

6 5

4

32

1

AP1

AP2AP3

AP4

AP5

AP6

Lab A

Lab B

Balcony

Figure 4 Reference point selection and graph construction

collection the localization serverwill determine the real-timeposition for each entity

32 Reference Point Selection and RSSI Sampling In thissubsection we first describe the rule for reference pointselection in the target field Inside the building the fixedstructures have formed the natural partitions such as officesmeeting rooms and aisles Generally a representationalposition will be selected in each area such as points 3 and 4in Figure 4 However there are some rooms with a larger sizeWewill choosemultiple reference points which are uniformlydistributed in one room such as points 10 and 11 in Figure 4As the RSSI information varies much more in a large roomthe selection ofmultiple reference pointsmay help to improvethe localization accuracy and efficiency After selection ofreference points we sample the RSSI information on eachreference point for a durative period such as 30 minutes andkeep the sampling results into the localization database on theserver To save the time multiple cards are used to sample theRSSI information on several reference points simultaneouslyFor a certain access point AP

119894 the average RSSI value from

the reference point 119895 is denoted by 120583119894119895 and its variance is 120590

119894119895

Note that in the underground mine environment theworking field is mostly a linear area If a long corridor isregarded as one area the relative distance from the entity to

APmight be large To solve this we divide the linear lanewayinto some ldquovirtual areasrdquo in which the center position willbe chosen as the reference point of this virtual area Thesparse division will decrease the localization accuracy andthe dense divisionmay result in serious localization jumpingthus decreasing the localization smoothness Given a longlaneway we deploy two access points with a distance of 100mOur testing shows that it is an efficient way to divide thislaneway with 100m into 5 virtual areas where the length ofeach area is 20m In this way the localization algorithm willsatisfy both accuracy and smoothness

33 Graph Construction for Reference Points In this sub-section a connected graph 119866 will be constructed for thereference points 119881 which is used to determine the track ofeach mobile entity and to improve the localization smooth-ness According to the description each reference point 119906

belongs to a certain partition or area denoted by119860(119906) In thefollowing 119906 and V are two reference points We describe therules for graph construction

(1) 119860(119906) and 119860(V) are two areas that is 119860(119906) = 119860(V)There are three cases to be discussed

(a) 119860(119906) and119860(V) are not connected or shared witha door Two points are not connected either

(b) 119860(119906) and 119860(V) are connected or shared with adoor Moreover points 119906 and V are the shortestpoint pair between two connected areas Twopoints are connected

(c) 119860(119906) and 119860(V) are connected or shared with adoor However points 119906 and V are not the short-est point pair between two connected areas Twopoints cannot be connected

(2) 119860(119906) and 119860(V) are the same area that is 119860(119906) =

119860(V) Two points are also connected in the graph 119866Generally if there are more than two reference pointsin one area they are all connected with others

In this way we have constructed an undirected graph119866 illustrated in Figure 4 By rule 1-a point 3 cannot beconnected with point 7 for two areas are not connected Byrule 1-b point 3 is connected with point 2 However point3 cannot be connected with point 1 according to rule 1-cFollowed by rule 2 point 10 and point 11 are connected forthey belong to the same area

34 Tree-Based Smooth Localization As described above thetime is divided into many discrete slots For example thelength of each time slot is 1 second During each time slotthemobile entity will locally broadcast the detectionmessageThe neighboring access points capture the detectionmessagecompute the RSSI values and transmit to the server As aresult the server can obtain the RSSI information (or RSSIvector) about each mobile entity in this time slot This vectorcan be expressed as (AP

1198941

RSSI1198941

) (AP119894119904

RSSI119894119904

) whereeach item (AP

119894119895

RSSI119894119895

) denotes that access point AP119894119895

has


detected the RSSI value of RSSI119894119895

from the mobile entity and119904 is the number of RSSI values in the vector

The previous works [8 10] have proposed a RSSI-distancemethod to measure the matching similarity between thecurrent RSSI vector and each sampled RSSI vector in thelocalization database However RSSI information betweentwo fixed nodes is not stable as time passed and wirelesstransmission is apt to unreliable which may lead to theincorrect localization result worse path matching errorand wrong jumping ratio Next we will design a tree-based smooth localization (TSL) mechanism for practicalindoor tracking This subalgorithm contains three stepsweight assignment accumulative weight computation andlocalization determination

In the first step each reference point will be assigned aweight by RSSI information For the mobile entity 119906 assumethat the RSSI information detected byAP

119894in current time slot

is denoted by 119909119894 It is also assumed that the RSSI information

obeys the Poisson distribution [11] Thus the probability thatthis entity locates on the position 119895 from the view of AP

119894is

119875119894119895=

1

120590119894119895radic2120587

119890minus(119909119894minus120583119894119895)221205902

119894119895 (3)

where the constants 120583119894119895and 120590

119894119895are introduced in the above

subsection The AP set that receives the RSSI informationfrom mobile entity 119906 in the time slot 119905 + 1 is denoted byAP1198941

AP1198942

AP119894119904

where 1198941 1198942 119894

119904isin [1119898] and 119898 is

the number of access points in the system The combinationprobability that the entity locates on the reference point 119895 isdenoted by 119875

119895as

119875119895=

119904

prod

119896=1

119875119894119896119895 (4)

With the gathered RSSI information during the time slot119905+1 we assign a weight119908(119895) for each reference point 119895 whichdenotes the possibility that the mobile entity 119906may locate onthis point with the collected RSSI information The weight ofeach area is computed as

119908 (119895) =

119875119895

sum119899

119896=1119875119896

(5)

where 119899 is the number of reference points in the system Bythis way each reference point has been assigned a weightNote that the weight assignment method is not unique Wealso can use another method for weight assignment Thoughthe weight is not strictly accurate this is useful to predict themoving track

The second step of TSL will compute the accumulativeweight for each reference point on the constructed tree Forsimplicity 119871(119905) denotes the localization result of time slot 119905for the mobile entity 119906 To determine the localization in timeslot 119905 + 1 the algorithm constructs a width-first searching(WFS) tree 119879 rooted at the point 119871(119905) for graph119866 Each point119895 knows its children set denoted by chl(119895) We compute theaccumulative weight for point 119895 denoted byAw(119895) as followsFor each leaf point in tree 119879 the accumulative weight is its

TSL sub-algorithmStep 1 Weight AssignmentFor each AP

119894isin AP

1198941AP1198942 AP

119894119904

For each reference point 119895119875119894119895=

1

120590119894119895radic2120587

119890minus(119909119894minus120583119894119895)

2(21205902

119894119895)

119875119895=

119904

prod

119896=1

119875119894119896119895

119908(119895) =

119875119895

sum119899

119896=1119875119896

Step 2 Accumulative Weight ComputationConstruct a tree 119879 rooted at 119871(119905) for graph 119866

For each reference point 119895

Aw(119895) =

119908(119895) 119895 is a leaf119908 (119895) + sum

119896isin119888ℎ119897(119895)

Aw(119896) else

Step 3 Localization Determination119867119900119901 = 0119901119900119904 = 119899119906119897119897For each point 119895 doIf Aw(119895) ge 119908

0

If ℎ119900119901119895gt 119867119900119901

119901119900119904 = 119895

119867119900119901 = ℎ119900119901119895

If 119867119900119901 gt 0

119871(119905 + 1) = 119901119900119904

Else119871(119905 + 1) = 119871(119905)

Algorithm 1 Formal description of the TSL subalgorithm

weight For each intermediate point the accumulative weightis the sum of accumulative weights of all the children nodesplus its weight That is

Aw (119895) =

119908(119895) 119895 is a leaf119908 (119895) + sum

119896isinchl(119895)Aw (119896) else (6)

Finally the TSL algorithmwill search for a reference pointwhose accumulative weight is more than a threshold 119908

0and

furthest from the root denoted by 1198950 Note that the distance

of two nodes is defined as the hop number between two nodesin the tree 119879 As a result the mobile entity will locate onthe point 119895

0in time slot 119905 + 1 That is 119871(119905 + 1) = 119895

0 This

indicates that the mobile entity has moved to the referencepoint 119895

0in time slot 119905 + 1 with very high probability If

there is no point whose accumulative weight is more than1199080 we regard that the tracked entity is static during time

slot 119905 + 1 That is 119871(119905 + 1) = 119871(119905) Though the entity maymove in this time slot the system cannot correctly judge themoving track as the RSSI information is unstableThe systemwill take a localization decision with a delay As a result thewrong localization jumping among several reference pointswill be avoided as possibleThe TSL subalgorithm is formallydescribed in Algorithm 1


2

1 3 4

6 8

7

Figure 5 Illustration of localization smooth

Table 1 An example of weights and accumulative weights

Point 1 3 4 6 7 8119882 010 010 015 020 010 025Aw 010 010 015 020 055 025

35 Illustration of the RSL Algorithm We illustrate the RSLalgorithm by an example The reference points are chosenand drawn in Figure 4 The parameter 119908

0is 050 For time

slot 119905 assume that 119871(119905) = 2 To compute the localizationof time slot 119905 + 1 the algorithm first constructs a WFS treerooted at point 2 shown in Figure 5 Assume that the mobileentity is currently located around the reference point 8 AfterRSSI collection wewill compute theweight and accumulativeweight of each point By Table 1 the accumulative weightsof point 6 and point 8 are 020 and 025 respectively Theaccumulative weight of point 7 is the sum of weights of points6 7 and 8 So Aw(7) = 020+010+025 = 055 According tothe RSSI information the algorithm may not distinguish thereal-time position either point 6 or point 8 for the mobileentity However the system can regard that the mobile entityhas moved to point 7 with very high probability We think itis reasonable for point 7 to lie on themoving path from point2 to the current position (ie point 8)

351 Algorithm Discussion When locating the mobile entityin each time slot the algorithm computes the weights andaccumulativeweights for all the reference points As the num-bers of reference points and mobile entities are both largethe requirement of computational capacity is very high forthe practical localization system To improve the computationefficiency we propose a local searching mechanism for theRSL algorithm Given a maximum velocity of the mobileentity each entitywill notmove a long distance in a short time(such as 1 s 2 s etc) For time slot 119905+1 the RSL algorithmwillconstruct a local tree 1198791015840 whose maximum depth is not morethan 119896 rooted at 119871(119905) where 119896 is predefined constant (eg 5or 10) in the system

4 Experimental Results

This section presents the numerical results to demonstratethe efficiency and smoothness of the proposed localization

algorithm Though there are some localization algorithmsbased on RF techniques they all require the additional con-ditions for indoor localization For example LANDMARCrequires the dense deployment of beacon nodes and the EZalgorithm [12] will occasionally fix the localization by GPSMoreover there is no special work on smooth localizationAs a result we mainly evaluate the performance of the RSLalgorithm by comparing with the RADAR system [8] onthe WiFi test platform The RADAR system introduces twodifferent methods of weight assignment Euclidean distanceand Manhattan distance So we will denote RAD + Euc andRAD + Man to express the localization methods with dif-ferent weight assignments The experiments mainly observethe performance of three localization measurements pointdecision accuracy path matching error and wrong jumpingratio respectively The definitions for these measurementsare described in the above section The accumulative weightthreshold 119908

0is set as 050 Besides this proposed algorithm

has little effect on the layout of the goods around the mobileentity regarding the spatial uniqueness of RSSI distribution

41 Experiment Environment The experiment is conductedat the Demo Center of Alcatel-Lucent Shanghai Bell Co LtdThere are totally 12 reference points and 6 access points in anarea of about 400 squares meters The reference points andAP deployments are also illustrated in Figure 4

42 Numerical Results for Localization Algorithms The firstexperiment mainly observes the performance of point deci-sion accuracy for different algorithms In particular themobile entity will statically locate in one place for continuous115 time slots in the evaluations By the collected RSSIinformation we can compute the point decision accuracyfor different algorithms Table 2 gives the PDA comparisonof different algorithms On the average the RSL algorithmimproves the PDA by about 15 and 8 compared to theRAD + Euc and RAD + Man algorithms Considering thehighlighted worst case RSL can enhance the worst PDA from054 to 069 Thus the proposed algorithm can get smootherlocalization compared to the RADAR system for the staticcase

The second experiment observes the performance ofpath matching error and wrong jumping ratio for differentalgorithms We select two different paths in the target fieldOne is 1-2-7-8-9-12 denoted by path A The other is 5-6-7-8-10-11 denoted by path BMoreover the evaluations adopt twodifferent moving patterns through each path One is movingwith the uniform velocity denoted by pattern a The other issimilar with case a except that the mobile entity will stay oneach reference point for 10 seconds denoted by pattern b Wegive the evaluation results in Table 3 to Table 6 in which thenumbers in the brackets denote the lasting time for differentpaths For pattern a the proposed algorithm reduces thepath matching error of about 92 compared to the RADARsystem Moreover the RSL algorithm decreases the wrongjumping ratio by at least 30 compared to theRADAR systemfromTables 3 and 4 For moving pattern b the RSL algorithmwill reduce the path matching error of about 151 compared


Table 2 PDA comparison for different algorithms

Number Real position RAD + Euc RAD +Man RSL1 1 091 091 0912 1 092 092 0943 1 068 081 0934 7 054 062 0715 7 077 077 0836 9 079 083 0877 9 071 070 0748 11 064 072 0849 11 055 067 069

Table 3 Performance for path A under pattern a (64 s)

Number Measurement RAD + Euc RAD +Man RSL1 PME 036 036 0302 WJR 2031 2188 938

Table 4 Performance for path B under pattern a (61 s)


Table 5 Performance for path A under pattern b (109 s)


Table 6 Performance for path B under pattern b (104 s)


with the RADAR system At the same time the proposedalgorithm also decreases the wrong jumping ratio by about65 compared with the RADAR system from Table 5 andTable 6 That is due to the tree-based mechanism adopted inRSL which helps to avoid the wrong jumps in localization asmuch as possible

Based on the evaluation results the proposed algorithmcan improve three localizationmeasurements compared withthe previous algorithms such as RAD + Euc and RAD +Man In particular the RSL algorithm will work excellentlyfor the performance of wrong jumping ratio As a result thisalgorithm improves the localization smoothness comparedwith the previous algorithms intuitively

5 Conclusion

In this paper a novel RSL algorithm is designed imple-mented and validated This algorithm purely depends onRF technique and uses a serial of access points to track the

mobile entities in the indoor environment The evaluationsdemonstrate that this system is more effective than theprevious related works On our evaluation the RSL algorithmcan reduce the path matching error and wrong jumpingratio by about 10 and 30 compared with the previoussystems As the RSSI information is unstable RSL may notbe fully smooth In the future our team will continue toimprove the point decision accuracy path matching errorand wrong jumping ratio In many applications delay isanother important measurement for localization In thefuture wewill study the tradeoff between the smoothness anddelay for real-time localization

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgment

This paper is sponsored by the Chinese National Science andTechnology Major Project 2012ZX03005009

References

[1] Cricket Projects ldquoCricket v2 user manualrdquo Tech Rep MITComputer Science and Artificial Intelligence Lab CambridgeMass USA 2004

[2] R Want A Hopper V Falcao and J Gibbons ldquoActive badgelocation systemrdquo ACM Transactions on Information Systemsvol 10 no 1 pp 91ndash102 1992

[3] N B Priyantha A Chakraborty and H Balakrishnan ldquoTheCricket location-support systemrdquo in Proceedings of the 6thAnnual ACM International Conference on Mobile Computingand Networking (MOBICOM rsquo00) pp 32ndash43 ACM Press NewYork NY USA

[4] N Patwari and A O Hero III ldquoUsing proximity and quantizedRSS for sensor localization in wireless networksrdquo in Proceedingsof the 2nd ACM International Workshop on Wireless SensorNetworks and Applications (WSNA rsquo03) pp 20ndash29 San DiegoCalif USA September 2003

[5] K Yedavalli B Krishnamachari S Ravulat and B SrinivasanldquoEcolocation a sequence based technique for RF localization inwireless sensor networksrdquo in Proceedings of the 4th International


Symposium on Information Processing in Sensor Networks (IPSNrsquo05) pp 285ndash292 April 2005

[6] X Shen ZWang P Jiang R Lin and Y Sun ldquoConnectivity andRSSI based localization scheme for wireless sensor networksrdquoin Proceedings of the International Conference on IntelligentComputing (ICIC rsquo05) pp 578ndash587 Hefei China August 2005

[7] Y Lee E Stuntebeck and S CMiller ldquoMERITMEsh of RF sen-sors for indoor trackingrdquo in Proceedings of the 3rd Annual IEEECommunications Society on Sensor andAdHocCommunicationsand Networks (Secon 06) pp 545ndash554 September 2006

[8] P Bahl and V N Padmanabhan ldquoRADAR an in-building RF-based user location and tracking systemrdquo in Proceedings ofthe 19th Annual Joint Conference of the IEEE Computer andCommunications Societies (IEEE INFOCOM rsquo00) vol 2 pp775ndash784 Tel Aviv Israel March 2000

[9] M N Lionel Y Liu Y C Lau and A P Patil ldquoLANDMARCIndoor location sensing using active RFIDrdquoWireless Networksvol 10 no 6 pp 701ndash710 2004

[10] M Terwilliger A Gupta V Bhuse Z H Kamal and M ASalahuddin ldquoA localization system using wireless networksensors a comparison of two techniquesrdquo in Proceedings ofthe Workshop on Positioning Navigation and Communication(WPNC 04) March 2004

[11] S Saha K Chaudhuri D Sanghi and P Bhagwat ldquoLocationdetermination of a mobile device using IEEE 80211b accesspoint signalsrdquo in Proceedings of the IEEE Wireless Communica-tions and Networking (WCNC rsquo03) vol 3 pp 1987ndash1992 2003

[12] K Chintalapudi A P Iyer and V N Padmanabhan ldquoIndoorlocalization without the painrdquo in Proceedings of the 16th AnnualConference on Mobile Computing and Networking (MobiComrsquo10) pp 173ndash184 September 2010

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



RSSI and Euclidean distance several works [4] compute thenodesrsquo localizations based on the connectivity informationThen RSSI is used to enhance the location precision [5 6]However there is mere work dealing with the unreliabil-ity and instability of RSSI which make many algorithmsunfit for the practical applications The existing RF-basedtracking systems mostly focus on the precise position oraverage position error However no apparent performancemeasurement has been studied for smooth localization Forexample the MERIT system [7] defines a new measurementfor indoor localization called area decision accuracy todetermine where the entity locates by comparing the averageRSSIs of different areas As RSSI is not always stable orreliable it does not work well under the indoor environmentMERIT [7] requires the denser AP deployment and doesnot fit for the areas with different area sizes As in the SAMapplication wrong area decision will result in wrong assetsearching and asset statistics With the negative features ofRSSI smooth localization becomes an extremely difficultchallenge for researchers As we know there is no specialwork aiming to smooth localization The main contributionof this paper is as follows

(i) Three novel measurements point decision accuracypath matching error and wrong jumping ratio aredefined to express the localization smoothness forindoor environment

(ii) A RSSI-based smooth localization (RSL) algorithm isproposed for the indoor localization The algorithmfirst selects some reference points and constructsa connected graph for these points based on thearchitectural features Then reference points will beassigned weights and accumulative weights by thecollectedRSSI information Finally a novel tree-basedmechanism is designed to estimate the moving trackwith high probability for smooth localization

(iii) The proposed algorithm is validated on the WiFi testbedThe evaluation results indicate that the proposedalgorithm can get better localization smoothnesscompared with the previous works For example theRSL algorithm can reduce the path matching errorand wrong jumping ratio by about 10 and 30 overthe RADAR system

In Section 2 the localization framework is introducedand three novel performance measurements for smoothlocalization are defined The RSSI-based smooth localizationalgorithm is discussed in detail in Section 3The validation isdescribed in Section 4 according to the above measurementsFinally the conclusion is drawn in Section 5

2 Localization Performance Measurement

21 Localization Framework Many algorithms of previousworkmay locate themobile entity to arbitrary positions in thefield However as RSSI information is unstable and wirelesstransmission apt to be unreliable the traditional RSSI-basedmethods might result in unstable localization That is thelocalization result will jump from one point to another while

AB

u1

u2

u3

Figure 1 Illustration of localization measurement

the entity is stationary to a fixed spot To provide smoothlocalization this procedure mainly consists of two steps(1) RSSI sampling and (2) position determination Like theprevious method [8] the localization framework first selectsmany reference points (also called representational points)in the field For example the central point of a room will beselected to represent this roomThe system samples the RSSIinformation on these discrete points for a durative periodIn the runtime the mobile entity will be located on somepresampled positions by the collected RSSI informationIn the second step different mechanisms are designed toimprove the localization performanceThemain advantage ofthis framework is to increase the localization efficiency andstability compared with the previous mechanisms [7] Thetask of this paper is to design efficient methods to improvethe localization smoothness under this practical framework

22 LocalizationMeasurement Definition Theprevious algo-rithms [5 7ndash9] mostly take the average position errorwhich denotes the distance between the real position andlocated position as the main performance measurementHowever this measurement cannot often reflect the realeffect for indoor localization For example there are tworooms in a field shown in Figure 1 Assume that the currentposition of a mobile entity is 119906

1 Two localization algorithms

obtain the different position results denoted by and throughRSSI information From the view of average position errorposition 119906

2is closer to 119906

1than to 119906

3 so that theoretically

the average position error between 1199061and 119906

2is smaller than

that of 1199061and 119906

3 However 119906

1is located in the same room as

1199063 Thus it is more reasonable to locate the mobile entity on

position 1199063compared to position 119906

2

In the following we will define some practical perfor-mance measurements for indoor localization There are twodifferent scenes for indoor localization One is static theother is mobile

For static scenes the entitymostly stays at one position oraround for a long time Under the above localization frame-work each position in the field will be logically appointed to areference point For example two points are in the same areaor close to each other We adopt the point decision accuracy(PDA) to measure the localization efficiency under the staticscenes In this case point decision accuracy is defined as thecorrect probability that the mobile entity is located at theappointed reference point by the localization algorithm Forexample as shown in Figure 1 we regard that each point in









119899 (1)


WJR =119898

119899times 100 (2)









2=









paths LP2and LP

3Though LP

2and LP

3reach the same PME


3for WJR







12

10

11

9

8

7

6 5

4

32

1

AP1

AP2AP3

AP4

AP5

AP6

Lab A

Lab B

Balcony






119894119895













1198941

RSSI1198941

) (AP119894119904

RSSI119894119904


119894119895

RSSI119894119895


has









119894is

119875119894119895=

1

120590119894119895radic2120587

119890minus(119909119894minus120583119894119895)221205902

119894119895 (3)




AP1198942

AP119894119904

where 1198941 1198942 119894

119904isin [1119898] and 119898 is


119895as

119875119895=

119904

prod

119896=1

119875119894119896119895 (4)


119908 (119895) =

119875119895

sum119899

119896=1119875119896

(5)




119894isin AP

1198941AP1198942 AP

119894119904


1

120590119894119895radic2120587

119890minus(119909119894minus120583119894119895)

2(21205902

119894119895)

119875119895=

119904

prod

119896=1

119875119894119896119895

119908(119895) =

119875119895

sum119899

119896=1119875119896



Aw(119895) =

119908(119895) 119895 is a leaf119908 (119895) + sum

119896isin119888ℎ119897(119895)

Aw(119896) else


0

If ℎ119900119901119895gt 119867119900119901

119901119900119904 = 119895

119867119900119901 = ℎ119900119901119895

If 119867119900119901 gt 0

119871(119905 + 1) = 119901119900119904

Else119871(119905 + 1) = 119871(119905)



Aw (119895) =

119908(119895) 119895 is a leaf119908 (119895) + sum

119896isinchl(119895)Aw (119896) else (6)


0and




0 This






2

1 3 4

6 8

7



Point 1 3 4 6 7 8119882 010 010 015 020 010 025Aw 010 010 015 020 055 025


0is 050 For time
























5 Conclusion





Acknowledgment


References

















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation

















119899 (1)


WJR =119898

119899times 100 (2)









2=









paths LP2and LP

3Though LP

2and LP

3reach the same PME


3for WJR







12

10

11

9

8

7

6 5

4

32

1

AP1

AP2AP3

AP4

AP5

AP6

Lab A

Lab B

Balcony






119894119895













1198941

RSSI1198941

) (AP119894119904

RSSI119894119904


119894119895

RSSI119894119895


has









119894is

119875119894119895=

1

120590119894119895radic2120587

119890minus(119909119894minus120583119894119895)221205902

119894119895 (3)




AP1198942

AP119894119904

where 1198941 1198942 119894

119904isin [1119898] and 119898 is


119895as

119875119895=

119904

prod

119896=1

119875119894119896119895 (4)


119908 (119895) =

119875119895

sum119899

119896=1119875119896

(5)




119894isin AP

1198941AP1198942 AP

119894119904


1

120590119894119895radic2120587

119890minus(119909119894minus120583119894119895)

2(21205902

119894119895)

119875119895=

119904

prod

119896=1

119875119894119896119895

119908(119895) =

119875119895

sum119899

119896=1119875119896



Aw(119895) =

119908(119895) 119895 is a leaf119908 (119895) + sum

119896isin119888ℎ119897(119895)

Aw(119896) else


0

If ℎ119900119901119895gt 119867119900119901

119901119900119904 = 119895

119867119900119901 = ℎ119900119901119895

If 119867119900119901 gt 0

119871(119905 + 1) = 119901119900119904

Else119871(119905 + 1) = 119871(119905)



Aw (119895) =

119908(119895) 119895 is a leaf119908 (119895) + sum

119896isinchl(119895)Aw (119896) else (6)


0and




0 This






2

1 3 4

6 8

7



Point 1 3 4 6 7 8119882 010 010 015 020 010 025Aw 010 010 015 020 055 025


0is 050 For time
























5 Conclusion





Acknowledgment


References

















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











12

10

11

9

8

7

6 5

4

32

1

AP1

AP2AP3

AP4

AP5

AP6

Lab A

Lab B

Balcony






119894119895













1198941

RSSI1198941

) (AP119894119904

RSSI119894119904


119894119895

RSSI119894119895


has









119894is

119875119894119895=

1

120590119894119895radic2120587

119890minus(119909119894minus120583119894119895)221205902

119894119895 (3)




AP1198942

AP119894119904

where 1198941 1198942 119894

119904isin [1119898] and 119898 is


119895as

119875119895=

119904

prod

119896=1

119875119894119896119895 (4)


119908 (119895) =

119875119895

sum119899

119896=1119875119896

(5)




119894isin AP

1198941AP1198942 AP

119894119904


1

120590119894119895radic2120587

119890minus(119909119894minus120583119894119895)

2(21205902

119894119895)

119875119895=

119904

prod

119896=1

119875119894119896119895

119908(119895) =

119875119895

sum119899

119896=1119875119896



Aw(119895) =

119908(119895) 119895 is a leaf119908 (119895) + sum

119896isin119888ℎ119897(119895)

Aw(119896) else


0

If ℎ119900119901119895gt 119867119900119901

119901119900119904 = 119895

119867119900119901 = ℎ119900119901119895

If 119867119900119901 gt 0

119871(119905 + 1) = 119901119900119904

Else119871(119905 + 1) = 119871(119905)



Aw (119895) =

119908(119895) 119895 is a leaf119908 (119895) + sum

119896isinchl(119895)Aw (119896) else (6)


0and




0 This






2

1 3 4

6 8

7



Point 1 3 4 6 7 8119882 010 010 015 020 010 025Aw 010 010 015 020 055 025


0is 050 For time
























5 Conclusion





Acknowledgment


References

















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation

















119894is

119875119894119895=

1

120590119894119895radic2120587

119890minus(119909119894minus120583119894119895)221205902

119894119895 (3)




AP1198942

AP119894119904

where 1198941 1198942 119894

119904isin [1119898] and 119898 is


119895as

119875119895=

119904

prod

119896=1

119875119894119896119895 (4)


119908 (119895) =

119875119895

sum119899

119896=1119875119896

(5)




119894isin AP

1198941AP1198942 AP

119894119904


1

120590119894119895radic2120587

119890minus(119909119894minus120583119894119895)

2(21205902

119894119895)

119875119895=

119904

prod

119896=1

119875119894119896119895

119908(119895) =

119875119895

sum119899

119896=1119875119896



Aw(119895) =

119908(119895) 119895 is a leaf119908 (119895) + sum

119896isin119888ℎ119897(119895)

Aw(119896) else


0

If ℎ119900119901119895gt 119867119900119901

119901119900119904 = 119895

119867119900119901 = ℎ119900119901119895

If 119867119900119901 gt 0

119871(119905 + 1) = 119901119900119904

Else119871(119905 + 1) = 119871(119905)



Aw (119895) =

119908(119895) 119895 is a leaf119908 (119895) + sum

119896isinchl(119895)Aw (119896) else (6)


0and




0 This






2

1 3 4

6 8

7



Point 1 3 4 6 7 8119882 010 010 015 020 010 025Aw 010 010 015 020 055 025


0is 050 For time
























5 Conclusion





Acknowledgment


References

















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










2

1 3 4

6 8

7



Point 1 3 4 6 7 8119882 010 010 015 020 010 025Aw 010 010 015 020 055 025


0is 050 For time
























5 Conclusion





Acknowledgment


References

















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation






















5 Conclusion





Acknowledgment


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











RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









RSSI-Based Smooth Localization for Indoor Environment

Documents