review article a survey of crowd sensing opportunistic...

Review ArticleA Survey of Crowd Sensing Opportunistic Signals forIndoor Localization

Ling Pei1 Min Zhang2 Danping Zou1 Ruizhi Chen3 and Yuwei Chen4

1Shanghai Key Laboratory of Navigation and Location-Based Services School of Electronic Information and Electrical EngineeringShanghai Jiao Tong University Shanghai 200240 China2Department of Computer Science School of Electronic Information and Electrical Engineering Shanghai Jiao Tong UniversityShanghai 200240 China3State Key Laboratory of Information Engineering in Surveying Mapping and Remote Sensing Wuhan UniversityWuhan 430079 China4Department of Remote Sensing and Photogrammetry Finnish Geospatial Research Institute National Land Survey of Finland02431 Masala Finland

Correspondence should be addressed to Ling Pei lingpeisjtueducn

Received 17 January 2016 Accepted 26 April 2016

Academic Editor Daniele Riboni

Copyright copy 2016 Ling Pei et alThis is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Sensor-rich smartphone enables a novel approach to training the fingerprint database for mobile indoor localization via crowdsensing In this survey we discuss the crowd sensing based mobile indoor localization in terms of foundational knowledge signalsof fingerprints trajectory of obtaining fingerprints indoor maps evolution of a fingerprint database positioning algorithms state-of-the-art solutions and challenges The survey concludes that the crowd sensing is a low cost solution of generating and updatingan organic fingerprint database Although the crowd sensing concept is widely accepted by the academic community in these yearsthere are a lot of unsolved problems which hinder the concept of transferring into a practical system We address the challengesand predict future trends in the end

1 Introduction

(1) Mobile Indoor Localization Location is an essential ele-ment of fast expanding modern information People usuallyspend over 90 of their daily lives indoors where themobile device for example smartphone is like a shadowinseparably sticking to users which greatly increases theinterests of mobile indoor localization for academia andindustry alike Global Navigation Satellites Systems (GNSS)largely enrich the localization capability of mobile devicesoutdoors However GNSS signals are designed for outdoorapplications originally which lowers or disables satellite-based localization technologies indoors because of the weaksignal or blocked signal in non-line-of-sight (NLOS) situa-tions To address localization in GNSS-degraded or deniedarea manifold technologies are extensively researched

The sensor-rich smartphone offers the potential of con-tinuous localization even though the localization infrastruc-tures are not available Typically the measurements from

smartphone built-in sensors for example accelerometermagnetometer and gyroscope can be fused to estimatethe smartphone carrierrsquos motion dynamic such as speedheading orientation ormotion states An algorithm namelypedestrian dead reckoning (PDR) can utilize the abovesmartphone dynamic information to locate a pedestrianin GNSS challenging environments [1ndash3] Phone camerais another potential sensor of smartphone-based localiza-tion [4 5] For instance Zhou et al [6] propose a visualSLAM (Simultaneous Localization and Mapping) algorithmto track userrsquos location and simultaneously build a 3D mapusing the structure line of a building Taking advantage ofthe widely integrated three-dimensional magnetometers insmartphones IndoorAtlas provides the cutting edge mag-netic fingerprint-based indoor localization [7] Opportunis-tic radio frequency (RF) based signals such as Wi-Fi [8ndash10] Bluetooth [11] Near Field Communication (NFC) RadioFrequency Identification (RFID) and cellular networks are

Hindawi Publishing CorporationMobile Information SystemsVolume 2016 Article ID 4041291 16 pageshttpdxdoiorg10115520164041291

2 Mobile Information Systems

Observable signals

Location

Fingerprint

Current measurements

Fingerprint DB

Feature matching

Pos

Learning phase Positioning phase

Observable vector

Sam

ples Feature vectors

Loca

tions∙ Extracted features

∙ Location

middot middot middotmiddot middot middot

middot middot middot



Figure 1 Flowchart of fingerprinting-based localization approach

pervasive providing the existing infrastructures RF-basedopportunistic signals have been widely applied for indoorlocalization for indoor environments [12]

(2) Crowd Sensing In 2005 with outsources fast growingcrowdsourcing has emerged as a new collaborative paradigmin which crowds of people can collaborate and accomplisha specific task The most famous crowdsourcing projectis the reCAPTCHA (Completely Automated Public Turingtest to Tell Computers and Humans Apart) which is themost applied technology of preventing malicious codes fromabusing web services [13] The project achieves such goal byrequesting the user input such as the distorted characterswhich can only be performed by human being so far Themost elegant design of the reCAPTCHA is to transformsuch human load into a useful application where the userinputs can be utilized to digitize old publications which arehardly recognized by optical character recognition (OCR)Normally userrsquos inputs are the distorted fragment scannedfrom those old publications The reCAPTCHA toolkit hadbeen deployed in more than 40000 websites With theinvolvements of crowd web users the toolkit had digitalizedover 440 million words by 2008 In results the archivesof New York Times over 150 years were unconsciouslyrecognized and digitalized by web users within a fewmonths

Skyhook a worldwide Wi-Fi positioning provider col-lects the access points and their locations contributed byparticipants via crowdsourcing [9] The contribution will beused to update the Wi-Fi database after the quality controlcenter verifies the crowdsourced information Similar tocrowdsourcing the terms such as participatory sensing [14]human computation [15] opportunistic sensing [16] crowdcomputing [17] social sensing [18] and crowd sensing [19]have a common idea which is to aggregate the collectiveintelligence of a crowd to achieve a goal with inexpensive costUnconscious participation is the profile which fundamentallydistinguishes crowd sensing from other crowdsourcing-like

solutions which makes crowd sensing an ideal approach togenerating fingerprints for indoor localization by amountof common mobile phone users instead of expert sitesurveyors

(3) Organization of This Paper In this survey we review thestate-of-the-art indoor localization research related to crowdsensing solutions The contents are structured as followsSection 2 introduces the foundation of crowd sensing indoorlocalization in high level Section 3 examines the possiblesignals for fingerprints of indoor localization Section 4surveys the methods for obtaining the walking trajectory ofa participant Section 5 discusses the types of map used forindoor localization Section 6 looks at how the fingerprintsorganically change thewidely applied positioning algorithmsare discussed in Section 7 Section 8 compares the state-of-the-art solutions published recent years Section 9 points outthe challenges of crowd sensing based indoor localizationand then Section 10 concludes by identifying open researchtopics and future research directions

2 Foundation of Crowd Sensing forIndoor Localization

Crowd sensing is a paradigm using crowd contributionto achieve a complex task which is perfectly suitable forfingerprinting-based indoor localization To generate a timelyfingerprint database also known as a radio map in theWi-Fi localization this needs regular site surveys with anexpert in the target area In the crowd sensing solution theprofessional site survey phase is supposed to be replaced withnonprofessional sensing using off-shelf mobile devices suchas smartphone

Figure 1 describes a typical flowchart of fingerprinting-based localization approach which consists of two phasesnamely learning phase and positioning phase respectivelyDuring the learning phase also known as training phase

Mobile Information Systems 3

FC1 FCn

FCi

IndoormapFP DB

PositioningengineOS observable signals

Positionfeatures


Clusteringfusion and DBorganic update

BackendFCs fingerprint collections

Positions + features

FrontendBuilt-in sensors

OS1(eg Wi-Fi)

OS2 (egBluetooth)

OS3 (egmagneticsignals)

OS4 (egacousticsignals)

OSn (eg cellular)

Figure 2 A scheme of crowd sensing based indoor localization approach

or site survey a mobile device scans the observable signalsaround a given location After collecting one or more observ-able signal samples at the location a set of features can belearned and extracted from the raw signal samples By asso-ciating the given locationwith the extracted features a finger-print is formed for this specific location Given a target areaa number of fingerprints with locations covering the wholearea are created These fingerprints then form the fingerprintdatabase which can be furtherly used for localization duringthe positioning phase Within the area covered by the finger-print database a tracking device with the same signal scan-ning capability can sense around to get the measurements atcurrent location Features generated from current measure-ments are matched to the feature vectors which are stored inthe fingerprint database beforehand As a result a fingerprintcontaining the best matched feature vector and associatedlocation is obtained The location of the best matched finger-print is the current position of the tracking device

The conventional site survey needs a dedicated procedureto accomplish the fingerprints collection in a target areaTwo types of learning methods namely static learning anddynamic learning are usually applied The static learningfulfills the offline learning phase by dividing the area ofinterest into normalized grids collecting observable signalvector at each reference point in a static way The locationinformation of reference points are provided by a surveyormanually inputting From the machine learning aspect ofview we define such method as supervised learning methodIn the dynamic learning a surveyor walks along the corridorsor open space continuously Given the start point and endpoint of a predefined walking path the locations of eachsample can be interpolated linearly with the timestampassisted Therefore a surveyor can collect the fingerprintsconstantly Since only start point end point andwalking path

are needed to generate the fingerprints along a path we alsoname this kind of method semisupervised learning [21]

In order to provide localization services in a specific areait is essential to generate the fingerprint database of thisarea according to the above description whereas sensing thesurroundings at all the predefined locations in the target areais needed for generating the fingerprint database which istedious task for a large-scale spaceNevertheless this dull taskcan be replaced with crowd sensing if a participant can scanthe environment and obtain his location at the same timewithout any user intervention Therefore participants canperform daily activities and sample the signals continuouslyin the meantime

21 Scheme of a Crowd Sensing Based Mobile Indoor Localiza-tion Approach Figure 2 presents a typical scheme of a crowdsensing based mobile indoor localization approach whichconsists of frontend and backend

The frontend is amobile device for example smartphonewhich is always carried by a participant and plays a role asfingerprint collector The frontend needs to take a snapshotof fingerprints by combining its trajectories with collectedsignals at each sampling epoch Using the built-in sensorssuch as accelerometers gyroscopes magnetometers barom-eters and camera integrated with the observable signalssometimes the frontend can estimate the trajectory of a userSection 4 discusses the trajectory estimation in detail

Meanwhile possible signals for example Wi-Fi Blue-tooth magnetic fields and cellular signals are collectedfor generating or updating the fingerprint database on thebackend In this survey the fingerprint is a generic termwhich means a fingerprint could be solely derived from Wi-Fi Bluetooth magnetic field acoustic signals and so on Afingerprint namely combined fingerprint can also consist


Table 1 Crowd sensing versus expert survey

Metric Crowd sensing Expert surveyTime consumption High MediumLabor cost Low HighTrajectory Unsupervised SupervisedData quality Low HighData volume Large SmallCoverage Scalable LimitedTimeliness High LowMobile device Heterogeneous DedicatedWireless connection Needed UnnecessaryCarrying mode Diverse FixedComputational complexity High LowTrustworthiness Low High

of more than one-category signals Section 3 surveys theopportunistic signals applied as a fingerprint

The backend is a data processer which maintains anorganic fingerprint database meanwhile provides the feed-back for the position request Crowd sensing frontendscontribute fingerprint collections with erroneous and uncer-tainties The incoming data from a crowd need to clusterand fuse to keep a healthy organic fingerprint database Wewill look at the popular algorithms for dealing with crowddata fusion in Section 6 Given the current observation of amobile device the Positioning Engine estimates the locationof this mobile device by positioning algorithms which will bediscussed in Section 7

In addition indoor map is used for generatingupdatingthe fingerprint database and aiding in indoor positioning

22 Crowd Sensing versus Expert Survey Crowd sensing isdifferent approach for generating the fingerprint databasefrom the conventional expert site survey In the crowd sensingapproach the comprehensive site survey is replaced with adhoc incremental collection from participants Nonprofes-sional mobile users are involved via a noncooperative modeThe participants sense surroundings and contribute theirmeasurements silently In order to compare crowd sensingand expert survey as listed in Table 1 we introduce the termsof time consumption labor cost reference obtainment dataquality data volume coverage timeliness mobile devicewireless connection carryingmode computational complex-ity and trustworthiness to evaluate two approaches

221 Time Consumption Time consumption in this paperis a metric used for counting the time for generating afingerprint database of a whole target area There are twotypes of fingerprinting-based positioning algorithms namelydeterministic and probabilistic algorithms respectively Con-ventional expert site survey of probabilistic fingerprintingneeds enough samples to estimate the signal distributionof a grid For instance Youssef et al collected 300 samplesat each reference point to estimate a histogram-based jointdistribution [22] Each sample took one second whichmeans that 5 minutes is needed for generating a fingerprint

of grid The researchers from the Helsinki Institute forInformation Technology and Ekahau [23] also collected 40samples for each grid Xiang et al [24] and we [11] used amodel-based signal-distribution training scheme to decreasedown the number of training samples The deterministicfingerprinting algorithm needs less samples for instanceRADAR [25] combined four samples into one fingerprintThe time consumption of either probabilistic or deterministicsolution is assessable and decided by the size of the targetarea the density of grids and the accuracy requirementHowever crowd sensing based fingerprint learning is anunpredicted process due to the uncertain crowd movementflow which increases the time consumption of fingerprintdatabase generation

222 Labor Cost The conventional expert site survey needsdedicated offline learning phase which means a certain oflabor cost is necessary for generating the fingerprint databaseFurthermore regular additional site survey is demandedto maintain an updated database In the crowd sensingapproach participants contribute data voluntarily whichsignificantly cuts down the labor cost of the fingerprintdatabase generation

223 Reference Obtainment Normally the site survey is asupervised learning process with predefined grids or war-driving paths which provide the references of fingerprintsHowever the ideally crowd sensing is an unsupervisedlearning approach which bypasses the need of expert sitesurvey in order to avoid the user intervention

224 Data Quality In the expert survey a professionalsurveyor performs a strict war-driving with a specific devicewhich guarantees the quality of acquired data On the otherside crowd sensing is a voluntary participation mode inwhich participants cannot commit the data quality

225 Data Volume The data volume is a term representingthe data quantity during the learning phase in this sectionThe scale of a target area the density of grid the accuracyrequirement and the sample rate decide the data volume inthe expert survey In order to achieve a useful fingerprintwith satisfied accuracy the samples amount is no less thanthat of expert survey Considering that crowd sensing is anoncooperative working mode the overlapped learning isunavoidable which increases the volume of learning data

226 Coverage The coverage of an expert survey is definedin a limited area where localization services are required Incontrary crowd sensing provides a scalable coverage whichis dependent on the movement of participants The coverageextends with the participant walking area expanding

227 Timeliness Timeliness in this paper is used to evaluatehow much a fingerprint database can represent the currentsignal environment After the initial fingerprint databaseis generated regular or irregular site surveys are requiredto maintain an updated database Using crowd sensing


approach frontends continuously contribute the sensingsignals which refresh the database frequently

228 Mobile Device In the expert survey the mobile deviceof a frontend is always dedicated andwell calibrated to ensurethe quality of fingerprint database The nature design of acrowd sensing does limit the frontend which leads to thediversity of mobile devices

229 Wireless Connection Because the expert survey is anoffline process the collected data can be stored locally andthen postprocess them Therefore communication connec-tion is not obligatory However in order to collect the sensingdata from distributed frontends wireless communication iscompulsive

2210 Carrying Mode In the expert survey the surveyorholds a mobile device strictly to eliminate the unexpectederrors due to the diverse carrying modes However crowdsensing participants carry a frontend arbitrarily which intro-duces the errors to the backend process

2211 Computational Complexity This term is used to char-acterize the difficulty of generating a fingerprint databaseExpert survey keeps low computational complexity by adedicated site survey However in the crowd sensing basedsolution a backend fuses a large number of sensing datafrom many frontends to achieve a robust fingerprintingThe heterogeneous devices unguaranteed data quality anddistributed system increase the computational complexity

2212 Trustworthiness The contribution from crowd sens-ing is hard to evaluate because less or none user interventionis required Except the information from low cost sensors andradio frequency modules users merely provide additionalmessagesTherefore the trustworthiness of the crowd sensingbased fingerprint learning approach is lower than that of theexpert survey

3 Opportunistic Signals

In general a type of signal can be used for fingerprinting-based localization if it has unique features at varying locationsand the unique features can be observed repeatedly and stablyduring a certain period The following opportunistic signalshave been already considered for generating fingerprints

31 Wi-Fi Today Wi-Fi networks are widely spread andfound in almost every public and private building Mostmobile devices also contain a Wi-Fi module To implementa positioning technique in a Wi-Fi network would thereforebe very cost effective Different researchers propose differentsolutions to the implementation problem and how the differ-ent difficulties can be taken care of Most of them suggest theuse of distance measurements using RSS values or the use ofRSS fingerprints This is because the RSSI function is alreadybuilt in and no extra hardware is needed

32 Bluetooth As Bluetooth can be found in almost everysmartphone today it is an interesting technology for indoor

positioning Compared to Wi-Fi infrastructure classicalBluetooth access points are not widely deployed whichdecreases the possibility of Bluetooth-based indoor localiza-tion Since the introduction of Bluetooth 40 or BluetoothLow Energy the implementation of Bluetooth in othermobile devices and sensors is probably going to increaseThe cheap and long life BLE module boosts the Bluetooth-based positioning via trilateration cell-ID or fingerprintingHowever Bluetooth-based fingerprints still suffer from thedynamic indoor environment because of the use of radiowavesThe variance of Bluetooth RSS is even higher than thatof Wi-Fi which decreases the stability of the fingerprints

33 Magnetic Field With the availability of embedded mag-netometer on smartphones a new fingerprinting approachbased on magnetic field has been proposed This approachis based on the hypothesis that in an indoor setting themagnetic field is highly nonuniform and the magnetic fieldfluctuations arise from both natural and man-made sourcesTherefore the abnormalities of themagnetic field can be usedas fingerprints for indoor localization While this approachshares a similar idea as Wi-Fi fingerprinting it certainlyhas several advantages compared to Wi-Fi (1) ubiquity andreliability (2) independence of the infrastructure and (3)power efficiency

34 Image Features Vision-based robot navigation usingonly a commercial off-the-shelf camera has been widelyresearched in recent years Smartphone with high resolu-tion camera brings new method of image-based indoorlocalization Images within a building are taken beforehandThen information such as image features correspondingcoordinates and viewing angles are generated and storedin the image fingerprint database in the learning phaseWhile in the positioning phase user takes a new picture andsearches the best match image from the fingerprint databasevia the image features and additional information Finally theuserrsquos current location is indicated with the correspondingcoordinates of the best matched image

35 Cellular Networks A large number of cellular towersacross populated areas enable cellular network signals servingas one of the most useful positioning sources Cell-IDtriangulation and trilateration are normally applied algo-rithms for cellular network based positioning both indoorsand outdoors In the density urban area non-light-of-signsignals decrease the performance of above methods RSS-based fingerprinting is an option for positioning in this caseHowever the RSSs of cellular towers at one location arenot stable because of the factors such as dynamics in theenvironment user effect user orientation and multipathpropagation in the indoor environments which also decreasethe performance of cellular network based fingerprinting

36 Ambient Light Ambient light exists anywhere anytimeeven the dim light can be considered as a special case ofambient light Ambient light sensors have been miniatureenough and commonly embedded in a smartphone whichcan detect the light intensity of environments The light


intensity is varying with the location because the buildingand objects in the building make the light feature uniqueat different positions Therefore ambient light based posi-tioning can use existing sensors in smartphones withoutextra infrastructure which represents a low cost positioningsolution [26] However the light changes over time whichmakes positioning difficult using the absolute light intensity

37 Ambient Sound The ambient sound has the uniqueand repeatable features associated with a specific locationFor instance public area contains noise in the backgroundversus private place that is quieter Taking time domain andfrequency domain into account the features extracted fromambient sound recorded in a room using a phone micro-phone can be used to identify one place from another Forexample SurroundSense [27] achieves an average accuracyof 87 with 51 test stores via ambience fingerprinting

4 Walking Trajectory

The above opportunistic signals need to be georeferenced inthe corresponding fingerprint database Hence the trajectoryof a participant sensing signals is demanded Smartphone-based PDR and SLAM are two candidates for obtaining thewalking trajectory in the crowd sensing approach

41 Pedestrian Dead Reckoning Pedestrian dead reckoning(PDR) is a relative localization method which determinesthe displacement and orientation change of a pedestrian overa step Step detection step length estimation and headingdetermination form a PDR algorithm Normally the accel-erations observed from accelerometers are utilized to detect astepThen step length can be estimated using the informationsuch as step frequency mean of acceleration and variance ofacceleration Finally heading determination can be achievedby fusing the data from gyroscopes accelerometers andmagnetometers

The location of a pedestrian can be propagated as followsin the PDR method

119909119896+1 = 119909119896 + SL119896 sin 120579119896

119910119896+1 = 119910119896 + SL119896 cos 120579119896(1)

where 119909119896 and 119910119896 are the coordinates in north and eastdirections SL119896 is the step length and 120579119896 is the heading attime 119896 From (1) it is shown that we can estimate the positionof the pedestrian at any time given an initial position thestep length and the heading of the pedestrian derived fromsensors Providing the radiomap or floorplan EKFor particlefilter is usually applied for fusing the PDR estimations andprior data [28]

42 SLAM In the case that fingerprint database is notavailable SLAM can be used for tracking a participantand sensing the signals around the participant meantimeSLAM is a standard mathematical framework for iterativelyoptimizing (1) the trajectory (sequence of poses) or dynamicsof a user based on the prediction of the motion model

and observations of the user (the observations could belandmarks images range measurements or radio frequencymeasurements) and (2) the position of landmark and the2D3Dmap itself SLAM has been widely applied in roboticsRecently increasing research induces the SLAM frameworkinto the radio map or magnetic map generation such as Wi-Fi SLAM [29] and MagSLAM [30]

Taking the noise of sensor measurements into accounta SLAM problem can be formulated as a probabilistic formAssuming that a user moving around in an unknown envi-ronment with a sequence states of X1119905 = 1199091 1199092 119909119905 theuser senses the environment to obtain the perceptions Z1119905 =1199111 1199112 119911119905 and acquire the odometry measurementsD1119905 = 1198891 1198892 119889119905 Solving the full SLAM problem needsestimating posterior probability of the userrsquos trajectory X1119905and the map M of the environment given all measurementsand an initial state 1199090 The posterior probability is denoted as

119901 (X1119905M | D1119905Z1119905 1199090) (2)

In the crowd sensing based fingerprint generation approachD1119905 can be estimated by PDR via smartphones M couldbe represented as fingerprints 1199090 is an arbitrary locationin the target area The SLAM schemes such as FastSLAM[31] GraphSLAM [32] GP-LVM SLAM [29] or DPSLAM[33] could easily be implemented to run in real time on asmartphone

5 Indoor Maps

Indoor map so known as floor plan contains the usefulinformation of a building and relationships between roomsspaces and other physical features which instruct users toobtain the layout of the building find the location of interestor navigate to the destination For the indoor navigationpurposes raster image and vector data are two widely usedtypes of indoor maps

51 Raster Map A raster map actually is a type of digitalimage which is represented by reducible and enlargeablepixels The pixel is the smallest individual unit of the rastermap and not able to describe the object independently Acombination of the pixels with different colors or gray scalecan represent the object as point line or area In orderto utilize raster map for indoor navigation the orientationscale and coordinate system have to be predefined Theorientation indicates the deviation against the north whichenables the azimuth reading to align the raster map Thescale here defines the length in physical space of eachpixel Therefore the travel distance in physic can be plottedcorrectly on the raster map given the coordinate systemand the origin point defined beforehand The pixel does nothave the semantic representation which makes the rasterimage merely as a background in the localization scenariosThe raster map is a handy resource for indoor localizationsince the buildings such as shopping malls airports or trainstations provide their indoor maps on the website or on-siteCurrently the floorplans based on raster image have been


widely applied in the user self-generate indoor navigationapplications such as IndoorAtlas [7]

52 Vector Map The vector map is an abstract map thatderives from the geographical features which are representedby vectors such as point polyline and polygon accordingto their geometrical shapes The point focuses on the spatialposition of an object the polyline shows the connections ofthe points and the polygon indicates the area covered by aclosed polyline

Since the vector is applied for expressing point polylineand polygon the vector map is easier to register scale andoverlap diverse sources than the raster map Furthermorevector map allows much more analysis capability especiallyfor indoor road network Paths of indoor environments canbe represented by polyline in the vectormapApolyline entitycontains the spatial position of the start point end point andthe length of the line which satisfies the needs of networkanalysis in indoor environments The computational geome-try algorithms can be easily applied to constrain the walkingpath of a participant in the crowd sensing approach using theroad network or the layout of vector maps [28] Popular vec-tor data formats include AutoCADDXF Shapefile developedby Esri Simple Features specified by the Open GeospatialConsortium andGeographyMarkup Language byOpenGIS

6 Organic Fingerprint

The organic fingerprint [34] is a code word describing theevolution of a fingerprint which grows and updates graduallyand naturally In order to maintain an organic fingerprintdatabase in a large space over time crowd sensing is the bestapproach However fusing the data sensed from a crowd is acomplex task

61 Data Fusion Problem Smartphones which offer a greatplatform to extend the existing web based crowdsourcingapplications to a larger contributing crowd provide a varietyof ways for data collecting based on the increasing sensingcapabilities [35] A key challenge here is how to deal withthe unknown reliability or trustworthiness of informationreported from the crowd The reasons for it are multifoldFirstly diverse smartphones and various sensors have differ-ent levels of accuracies Secondly the quality of data cannotbe guaranteed since participants do not have the obligationto ensure the data quality unless the participants are paidTherefore the unreliability problem of data fusion risesunder the circumstance where multiple reports for the samesituation must be fused together

62 Data Fusion Solutions Recently a number of researchersproposed various methods [36ndash39] to estimate the reliabilityof the reports and compute their aggregated output In par-ticular many existing researches mostly in machine learningmainly focus on fusing multiple single-value observationscombined with the assessment of a userrsquos trustworthinessBachrach et al [40] proposed Crowd IQ which is a qualitymeasure of decisions based on aggregating opinions and

quantifies individual and crowd performances under thesame scale Their idea is to aggregate response IQ ques-tionnaire based on simple major voting mechanism mixedwith probabilistic graphical model-based machine learningapproach Kamar et al [41] constructed a set of Bayesianpredictive models within a crowdsourcing framework andalso employ multiple inferences to guide the selection andschedule the workers so as to maximize the overall efficiencyof large-scale crowdsourcing process Welinder et al [42]mainly deal with the image labelling problemThey proposeda way to estimate the underlying value (eg the class) of eachimage from (noisy) annotations provided by multiple anno-tators which is based on the image formation and annotationprocess In their work commonwisdom is to collect multiplelabels for each sample and adopt ldquomajor voterdquo to decide onthe correct labels In the worksmentioned above the primarymechanism in aggregating different opinions is ldquomajor voterdquowhich is widely used for centuries in almost everywherein peoplersquos daily life politics and so forth Whitehill et al[43] also proposed a probabilistic model to simultaneouslyinfer the label of each image An interesting point theyposed is that their model outperforms the common ldquomajorvoterdquo mechanism in inferring the labelsTheir work providedresearchers later on with a hint that ldquomajor voterdquo might notbe optimal in aggregating crowdsourced information thoughits simplicity makes it easy to implement

If we turn our eyesight to research in the field of mobilecomputing a similar problemofmultisensor fusionwill ariseA vast literature has addressed how to integrate multisensorestimates into one single output like covariance intersection[44] covariance union [45] and so forth The limitationof such problems is that they typically fuse the estimateswithout modeling the trustworthiness of the users or theyonly identify the unreliable estimates by some simple outlierdetection methods like kNN [46] spatial weighted outlierdetection (SOD) [47] local outlier factor (LOF) [48] and soforth The underlying assumption of these methods is thatthe noise in the data is only introduced by uncalibrated orfaulty sensors And thus an underlying problem is that theuntrustworthy information introduced by the crowd is nottaken into consideration in these methods

Park et al [34] proposed the Voronoi regions for convey-ing uncertainty and reasoning about gaps in coverage and aclustering method for identifying potentially erroneous userdata Users are requested to input to improve either coverageor accuracy Erroneous bind detection method is applied byclustering in signal space using linkage function In the year2013 Venanzi et al introduced the idea of learning the trustof the contributors which construct a likelihood model ofthe usersrsquo trustworthiness by scaling the uncertainty of itsmultiple estimates with trustworthiness parameters [49]Thiswork gives a framework for data fusion for crowdsourcingapplications

7 Fingerprinting-BasedPositioning Algorithms

As long as the fingerprint database is generatedmanifold pos-itioning algorithms can be applied according to application


requirements for instance deterministic approach like kNNapplied by RADAR [25] and probabilistic approaches usingBayesian theorem [22] By combining the other sensor infor-mation or floor plan the positioning solution can furtherapply the scheme such as EKF particle filter or SLAM

71 Deterministic Approach Thedeterministic fingerprintingapproach is actually a process of supervised learning andprediction The problem can be stated as follows given anunknown function thatmaps observations to locations alongwith training observable samples which can represent theactual distribution of observations produce an approximatefunction that is as close as possible to the actual mappingfunction In the learning step observation 119874119894119895 is the signalmeasured in location 119894 therefore the observable vectorR119894 canbe denoted as the following matrix

R119894 =

11987411 sdot sdot sdot 1198741119896

d

1198741198991 sdot sdot sdot 119874119899119896

(3)

where 119899 is the number of samples and 119896 is the number ofsignal sources Each column wraps the samples of one type ofsignal sources The manifold features can be extracted fromeach column to generate the fingerprint as

R119894 = []1198941 ]119894119902] (4)

where R119894 is the fingerprints of location 119894 and 119902 is the number ofextracted featuresThe pattern vector for locations is denotedas P = [

R1 R119898] where 119898 is the number of referencepoints Let L = [X1 X119898] denote the locations of all thereference points where the coordinates of reference point119894 is X119894 = 119909119894 119910119894 119911119894 Then the fingerprint database can beexpressed as

F =

[

[

[

[

[

X1 R1

X119898 R119898

]

]

]

]

]

(5)

In the prediction step the location of a smartphone canbe estimated by comparing the feature vector R119888 derivedfrom current observations with pattern vectors stored inthe fingerprint database The merits of such similarity areutilized for searching the nearest vector in the feature spaceThe comparison is based on distances in signal spaceThe distances such Euclidean distance Hamming distanceMahalanobis distance and Manhattan distance [50] areusually used for evaluating the similarity For instance in thekNN based deterministic algorithm the Euclidian distancecan be written as

119889 (R119888 R119894) =

10038171003817100381710038171003817

R119888 minus R11989410038171003817100381710038171003817 (6)

Finding the nearest neighbor equals searching the signalpatterns R119894 in the fingerprint database with the shortestsignal distance Then as shown in the following equation

the corresponding location 119897(R119888) associated with the signalpattern R119894 is the location we estimated

119897 (R119888) = argmin

R119894isinP119889 (

R119888 R119894) (7)

In order to improve the robustness the kNN algorithm takesthe 119896 nearest neighbors into account to estimate the finallocation 119909(R119888) as

119909 (R119888) =

1

119896

119896

sum

119894=1

119897119894 (R119888) (8)

where 119897119894(R119888) is the location associated with one of the nearestneighbors in signal domain

72 Probabilistic Approaches Compared to deterministicapproaches probabilistic approaches have higher accuracyand lower computational cost At each reference point thesignal probability distributions of all sources are stored If wedenote the fingerprint for the 119894th reference point as R119894 thenwe have

R119894

=

[

[

[

[

[

[

[

119875 (1198781 1198741 | X119894) 119875 (1198782 1198741 | X119894) sdot sdot sdot 119875 (119878119896 1198741 | X119894)119875 (1198781 1198742 | X119894) 119875 (1198782 1198742 | X119894) sdot sdot sdot 119875 (119878119896 1198742 | X119894)

d

119875 (1198781 119874V | X119894) 119875 (1198782 119874V | X119894) sdot sdot sdot 119875 (119878119896 119874V | X119894)

]

]

]

]

]

]

]

(9)

where 119878 stands for the signal source while 119874 refers to theobservation 119875(119878119896 119874V | X119894) is the probability of observedmeasurement 119874V from signal source 119878119896 given location X119894If this probability is calculated by counting the frequencyof certain observation occurred at a specific location wename it as nonparametric distribution that is histogramdistribution On the other hand if the probability is approx-imated by some distributions such as Gaussian distributionand Weibull distribution the parameters which can repre-sent the specific distribution are needed Therefore we callit as parametric distribution The main advantage of thenonparametric technique is the efficiency of calculating thelocation estimate while the parametric technique reducesthe fingerprint database size and smooths the distributionshape which leads to a slight computational advantage of theparametric technique over the nonparametric technique

Since the location is attached in the fingerprint R119894 thusfingerprint database can be expressed as

F = [R1R2 R119908] (10)

Providing the fingerprint database manifold probabilisticpositioning algorithms can be applied using the Bayesian the-orem such asMaximumLikelihood (ML) andMinimizationof Expected (distance) Error (MEE) The difference betweenthem is that ML always returns the location belonging to thereference point set of the fingerprint database while MEEalgorithm interpolates among the reference points In this


Others Signals of opportunity

WLAN

SensorsCamera

Accelerometer

Gyroscope

Digital compass Motion recognition

DRPDR techniques

Vision-based techniques

Calibration

(S O A B 120587)

Hidden Markov models

Output

Extra absolute positioning

Knowledgedatabase

GNSS RFID

Hybrid positioningalgorithms

Absolute positionvelocity and heading

Motion dynamicsinformation

Initializing

Integrity

Figure 3 The general high-level architecture of the HMM solution that fuses the measurements of the sensors and WLAN to estimateabsolution positions [20]

survey we take the Histogram-Based Maximum Likelihoodalgorithm as an example to explain the probabilistic position-ing approach [51]

Given the observation vector O = 1198781 1198741 1198782 1198742

119878119896 119874119896 from signal sources 1198781 to 119878119896 the problem is to findthe location X with the conditional probability 119875(X | O)

being maximized Using the Bayesian theorem

argmaxX [119875 (X | O)] = argmaxX [

119875 (O | X) 119875 (X)119875 (O)

] (11)

where119875(O) is constant for allX therefore (11) can be reducedas

argmaxX [119875 (X | O)] = argmax119897 [119875 (O | X) 119875 (X)] (12)

We assume that the mobile device has equal probability toaccess each reference point so 119875(X) can be considered asconstant in this case (12) can be simplified as

argmaxX [119875 (X | O)] = argmaxX [119875 (O | X)] (13)

Now it becomes a problem of finding the maximum condi-tional probability of

119875 (O | X) =119896

prod

119899=1

119875 (119878119899 119874119899 | X) (14)

where the conditional probability 119875(119878119899 119874119899 | X) is derivedfrom the histogram distribution prestored in the fingerprintdatabase

73 Hybrid Solutions The basic fingerprinting-based indoorlocalization algorithms such as kNN and probabilistic meth-ods will introduce location jitters because the original finger-printing algorithms do not take the motion dynamic modelinto account In order to achieve reliable indoor localizationhybrid solutions using both fingerprints and motion sensorsare widely adopted [20 52 53]

The potential fusion techniques include Kalman filterthe hidden Markov model and particle filter Kalman filter

is a common algorithm of multisources fusion which hasbeen extensively discussed in previous literatures Since themovement of a pedestrian is usually nonlinear trajectory anextended Kalman filter (EKF) is widely employed in whichthe nonlinearity can be dealt with by a Taylor expansionWhen the state transition and measurement models that isthe prediction and measurement update matrices are highlynonlinear the EKF gives particularly poor performancebecause the covariance is propagated through linearizationof the underlying nonlinear model [54] In this survey weintroduce HMM and particle filter based hybrid indoorlocalization approaches

In order to mitigate the impact of Wi-Fi fingerprintingcaused by RSSI variances Liu et al [20] proposed a HMM-based fusion framework as shown in Figure 3 to augmentthe Wi-Fi positioning by motion information In the HMMapproach a userrsquos positions are the hidden states to beestimated and the sequence of positions has the Markovproperty Observables in [20] are Wi-Fi RSSI and theemission probabilities of observables are probabilistic RSSI-position dependency obtained from a knowledge databaseThe accurate state transition probabilities can improve thelocalization results using the HMM approach

Particle filters are sequential Monte Carlo methods basedon point mass (or ldquoparticlerdquo) representations of probabilitydensities which can be applied to any state-space time-seriesmodel The state vector contains the kinematic informationof a pedestrian in the localization system The measurementvector represents noisy observations such as movementsderived from accelerometers gyroscopes and magnetome-ters and location estimated by signal fingerprinting [2854] The state vector can handle multivariate data andnonlinearnon-Gaussian processes

Figure 4 presents an approach which integrates stateupdates from PDR fingerprints and constraints from afloorplan to acquire the posterior distribution of a pedes-trianrsquos location [28] Particles wrap the position coordinatesheading parameters of step length and the weights derived


Sign

al st

reng

th (d

B)

minus50

minus70minus80minus90

minus60

87 6 5 4 3 2 1 0

876543210

Y (m)X (m)

Locations

Initializationcalibration

GPS

Mag

Wi-Fi

Gyro

AccMotion sensors Particle filter

Particle Ini

Particle Upt

ResamplingError

est

Floor mapfingerprint database

Stepdetection

Step lengthestimation

Headingestimation

Locationest

PDR

+Y

minusY

minusZ

+Z

minusX

+X

RawData

Motion recognition

Figure 4 A particle filter based hybrid indoor localization

from fingerprinting Besides the PDR parameters can also belearned and corrected during the particle propagation

8 The State-of-the-Art Solutions

81 Redpin [55] Redpin is one of the earliest signal basedindoor localization solutions which proposes to incorporateuser participation to build fingerprints rather than dependingon designated and time-consuming training process Redpindeveloped an adaptive indoor localization system involvingGSM Wi-Fi and Bluetooth signals Users could contributewithout much effort while at the same time guarantee room-level accuracy The Redpin system consists of two compo-nents the Sniffing component is designed to gather variouswireless signals in range to build fingerprints and the Locatorcomponent contains algorithm to locate a user using distancein signal domain User interacts with Redpin in the followingway after sniffing process if a user could be located by thesystem with the signal measurement heshe uploaded theuser will be informed of hisher current location otherwisethe user will be prompted to name hisher current locationThe performance of the system was evaluated by conductinglocalization experiment with 10 rooms and 9 of the roomswere recognized correctly in result whichmeans an accuracyof about 90

82 OIL [34] OIL targeted at organic room-level localizationto achieve which users need to integrate with OIL system tomake binds for rooms and correspondingWi-Fi fingerprintsIn [35] the authors mainly investigate the user promptingalgorithms in case that improper algorithm frustrates usersThey devised a user prompting algorithm based on VoronoiDiagram By arranging the spaces of interest into VoronoiDiagram they introduced a Spatial Uncertainty conceptwhich relates bounded regions with unbounded regions anddesign user prompting algorithm on top of this They also

considered the error binds filtering problem and proposedto use clustering in RSS signal space to eliminate wrongbinds To evaluate their model they conducted experimentsin a nine-story building with about 1400 spaces and with 19participants Over several days the mean error between thecentroid of estimated space and the centroid of ground truthroom decreases to less than 45m

83 WiFi-SLAM [29] WiFi-SLAM takes the initiatives tointegrate wireless signals with SLAM solutions to enable Wi-Fi localization without much training effort The authorspropose to use Gaussian Process Latent Variable Model (GP-LVM) in combination with a motion dynamics model todiscover the latent-space locations of unlabeledWi-Fi RSS Intheir likelihoodmodel of GP-LVM three types of constraintsare considered The locations rarr signal strength constraint iscaptured by the GP part which means that similar locationsshould have similar signals The motion dynamics part cap-tures the location rarr location constraints The last constraintsignal strength rarr location is a back constraint that is notprovided by GP-LVM and thus is implemented as a smoothinternalmapping An Isomapwhich could recover the overallstructure of Wi-Fi traces is used to generate acceptableinitialization for the optimization of whole GP-LVM modelTheir experiment reports a mean localization error of 397 plusmn

059meters

84 Zee [56] Zee is a zero-effort crowdsourcing indoorlocalization system which runs in the background on amobile device Specifically it requires no user-specific knowl-edge such as usersrsquo initial location stride length and phoneplacement It utilizes inertial sensors to track users whenthey traverse a path while simultaneously collecting Wi-Fisignals Initially a uniformdistribution overwhole floor placeis assumed for the initial location of the first user then bytracking the shape the user traverse and combining it with


the floor plan probabilities are eliminated and the predictedlocation converges to the ground truth one also backwardbelief propagation is leveraged to recover the whole pathThe following users work almost the same way as previousone except that their initial position distribution is narroweddown to a smaller region thanks to the Wi-Fi fingerprintcontributed by prior walks An augmented particle filter isapplied during the Wi-Fi crowdsourcing phase and then thedeterministic or probabilistic positioning algorithms can usethe Zee-based crowdsourcing fingerprint database Perfor-mance is evaluated by conducting experiments in a 35m by65m office buildingThe result shows that 50 of localizationerror is less than 12m and that 80 is less than 23m whichis lower than that of pure probabilistic positioning approachbut the site survey efforts are significantly reduced

85 LiFS [57] The authors of LiFS propose a novel frame-work for fingerprint-based indoor localization utilizingMDS(multidimensional scaling) twice tomap scanned RSS signalsto the path that a participant traversed Unlike previousSLAM based solutions LiFS only measures walking stepsbetween fingerprints thus avoiding dealing with long-termdrift of dead reckoning The first-time MDS is used is tomap the sample locations in real floor plan into a stress-free floor plan in which the Euclidean distance between twopositions reflects the walking distance of the correspondingpositions in real floor plan Then MDS is applied again togenerate the fingerprint space Reference points like corridorsand doors are recognized in fingerprint space and aremappedto their locations in the stress-free floor map Eventually allfingerprints can be associated with their corresponding loca-tions by performing a linear transformationThe localizationexperiment using RADAR-like algorithm yields a result of588m average localization error and 1091 room error ratein a 1600m2 experiment environment

86 MagSLAM [30] MagSLAM is a variation on SLAM(Simultaneous Localization and Mapping) which incorpo-rates ambient magnetic field signal In this framework themagnetic environment map which is generated from mag-netic field measurement is incorporated to build a DynamicBayesian Network (DBN) model that is extended fromFootSLAM [58] which utilizes pure odometry data Alsothe authors extend the spatially binned map in FootSLAMto a hierarchical way with different sized hexagonal cells toachieve an effective map representation On top of that asimple Monte Carlo approximation is applied to the resultsgenerated from the Bayesian estimator They presented theresults of 5 experiments with ground truth datasets compar-ing the performance under different settings of map layersand SLAMalgorithmusedTheir result shows thatMagSLAMcan achieve a localization accuracy of 9 cm to 22 cm whichgreatly exceeds the performance of using givenmagneticmapin the same environment

87 HiMLoc [59] HiMLoc is a hybrid framework that com-bines pedestrian dead reckoning (PDR) Wi-Fi fingerprint-ing and activity recognition to address crowdsourced indoor

positioning It also uses a particle filter to integrate the loca-tion estimation of activity classifier PDR Map Knowledgeand Wi-Fi positioning components The Wi-Fi fingerprintdatabase is then updated with the Wi-Fi observation and itscorresponding location annotation The performance of thisframework is evaluated in different scenarios single floormultiple floors and a new environment during deploymentIn most cases of the first two scenarios HiMLoc reportsa median accuracy of less than 3m When applied to newenvironment the performance of HiMLoc improves overtime due to the fast accuracy convergence which enables itto be easily deployed in new environment

88 UnLoc [60] The authors of UnLoc designed the unsu-pervised indoor localization framework based on the obser-vation that some positions in indoor environment bearsome characteristics that enable them to be identified Suchpositions are discovered by them in two phases and arethus categorized as Seed Landmarks andOrganic LandmarksSuch landmarks are leveraged to calibrate the pedestrianlocation at a landmark PDR drift can be reset while onelandmark is observed Deterministic algorithm is appliedfor matching a landmark War-driving is not necessaryneither are floorplans the system simultaneously computesthe locations of users and landmarks in a manner that theyconverge reasonably quickly They conducted experiments inthree different indoor buildings and yielded a result of 169mmean error

89 SmartSLAM [61] SmartSLAM is an indoor position-ing schema that switches between four different operatingregimes according to the prior knowledge it has about thespecific environmentThese four different methods are PDR-only EKF FEKFSLAM and DPSLAM respectively FEKF isan extension of particle filter applied on PDR incorporating aprior fingerprint map and signal measurement in the updatestage of an extended Kalman filter A FEKFSLAM is appliedwhen the prior fingerprint map is not available but PDRparameters are known to the system In this scheme theauthors build a novel empirical measurement model for loopclosure that captures the linear relationship between spatialseparation and fingerprintsrsquo Euclidean distance The systemwill turn to DPSLAM if the building floor plan is available orwhen the previous mentioned algorithms show bad perfor-mance DPSLAM uses a particle filter PDR fingerprintingas well as magnetic measurements and is thus more costlyGenerally a decision tree is utilized for transitions betweendifferent regimes to bring down the cost as much as possiblewhile at the same time guarantee poisoning accuracy in asmart way Experiments were conducted to evaluate the fourdifferent schemes the DPSLAM reports an accuracy of 16mwith 66 confidence and 27m with 95 confidence

810 FreeLoc [62] Themain goal of FreeLoc is to investigatehow to achieve efficient WiFi-based localization in an envi-ronment where device heterogeneity and multiple surveyorsexist To address these issues the authors devised a novel Key-Value fingerprint data structure with a parameter 120575 whereKey denotes a specific BSSID and Value is a vector containing


Table 2 Comparison table of state-of-the-art solutions

System Signals Frontend Algorithm Accuracy Participant Scale Placement Publish dateRedpin W B C MP DA 90 CR 10 10 RMs Free Sept 2008OIL W MP PA le45m 19 1400 RMs Free June 2010WiFi-SLAM W PDA PA 397 plusmn 059m NA 250ndash500m H Jan 2007Zee W SP PF le23m (80) NA 2275m2 Free Aug 2012LiFS W SP DA 588m 4 1600m2 H Aug 2012MagSLAM M XSens PF SLAM 9 cmndash22 cm NA 5 Bldgs Shoes Oct 2013HiMLoc W SP PF lt3m NA 600m2 HP Oct 2013UnLoc W M SP DA PDR 169m NA 3 bldgs HP June 2012

SmartSLAM W M SP PDR KF PFSLAM 27m (95) NA 600m2 NA Sept 2013

FreeLoc W SP DAlt2m

(Hallway)lt4m (Lab)

15 70 points Free April 2013

Elekspot W SPiPod DA PA 9187 CR 8 3 bldgs Free July 2012WicLoc W SP DA 465m 17 1600m2 NA June 2015Abbreviations in the table are list as follows(1) C Cellular network W Wi-Fi B Bluetooth M Magnetic field(2) MP Mobile Phone SP Smartphone PDA Personal Digital Assistant(3) DA Deterministic Approach PA Probabilistic Approaches PF Particle Filter KF Kalman Filter(4) RM Room Bldg Building(5) H Hand-held P Pocket(6) CR Correct Rate

BSSIDs of which RSS is 120575 weaker than the Key This relativerepresentation of RSS from APs along with 120575 not only makesthe system immune to device diversity but also increasesimilarity between fingerprints collected at slightly differentplace which enable merging Value factors for the same Keyunder multisurveyor circumstances Wi-Fi fingerprint datawas gathered at about 70 different locations in a building with4 different devices The result shows that cross device error isless than 2m for hallway 4m for laboratory

811 Elekspot [63] Elekspot is a platform that enables urbanindoor environment localization via crowdsourcing Thesystem is designed to support several major issues (inevitableproblems) in crowdsourcing framework system scalabilitydevice heterogeneity and robustness of lack of contributionA different method is proposed to deal with each of thesedesign goals respectively Specifically amethod named SSBI-n which makes inverted index for only BSSIDs with top nstrongest RSS strength instead of all BSSIDs in fingerprint isintroduced to reduce time in retrieving too many fingerprintand thus enable scalability To support device diversity theauthors propose to obtain linear relations between finger-prints from different devices automatically based on contri-butions in the same location and keep updating them Finallythey suggest using confidence value to denote reliabilityinstead of position error distance

812 WicLoc [64] WicLoc is an indoor crowdsourcing Wi-Fi fingerprinting framework which is based on a modifiedversion of MDS (multidimensional scaling) In their workthe authors generate distance matrix of fingerprints andtransform the distances into high-dimensional space through

MDS algorithm Furthermore they propose to use a certainnumber of anchor points to calibrate the output from classicalMDS algorithm Such anchor points are chosen from turningpoints near doors and corridors Experiments are conductedin an indoor area of about 1600m2 to evaluate their modeland two comparative models LiFS and EZ The result showsthat it achieves a mean localization error of 465m which issmaller than that of LiFS and EZ

813 Comparison of the State-of-the-Art Solutions As sum-marized in Table 2 we compare the above state-of-the artsolutions in terms of applied signals frontend type algo-rithms of generating fingerprints and positioning position-ing accuracy the number of participants in a crowd the scaleof field test the placement of frontend device and the pub-lished date of the researchThe accuracy reported by solutionslisted in Table 2 indicates the mean error of positioning inmeters or the rate of correct prediction in a percentage

Wi-Fi is the most adopted signal for crowd sensing dueto the existing infrastructure Magnetic field is the secondoption because of the free-infrastructure capability How-ever the lower-dimensional features of the magnetic fieldintroduce the ambiguity while positioning Handheld or in-pocket smartphone firmly takes the first order of the devicesapplied in crowd sensing even though the foot-mountedIMU such as Xsens has higher performance Deterministicor probabilistic fingerprinting and PDR are integrated withthe fusion algorithms such as Kalman filter particle filter orSLAM to achieve an accuracy of 1ndash6 meters Foot-mountedsolution is even higher in terms of accuracy The number ofthe crowd sensing participants and the scale of employed areaare limited in all the above systems


9 Challenges

Crowd sensing is an emerging solution for indoor localizationusing a smartphone However issues such as device diversityquality control carrying mode of a smartphone powerconsumption low cost of sensors high-dimensional dataparticipation willingness and privacy protection introducechallenges to achieve robust positioning results using crowdsensing fingerprint database

91 Device Diversity Diversiform smartphones indicateheterogeneous modules or sensors which are integratedinto phones with different smartphone manufacturers Forinstance inertial sensors with different performances willlead to different step detection thresholds Wi-Fi modulesfrom different providers have varying receive signal gainswhich make the RSSI varies using different devices at thesame location Finally device diversity will impact on bothlearning and positioning phases Although the Spearmanrank distance [65] can mitigate the effects of device diversityin the deterministic approaches such as kNN it is still achallenge in the probabilistic approaches

92 Quality Control Crowd sensing highly relies on theparticipant contribution in user intervention is demanded aslittle as possible Furthermore participants will not guaranteethe data quality unless they have commitments Thereforethe quality control on the frontend is essentially important torestrict the data before entering the backend Then furtherquality control is also needed on the backend However dataquality controls on both frontend and backend are rarelydiscussed in the state-of-the-art literatures

93 Unconstrained Mobility Less restriction or interventionis an important element which encourages the user toparticipate in the data contribution which means that theparticipant mobility should be unconstrained However thealgorithm such as PDR is highly relevant to the carryingmodeof a smartphone and the motion states of the user Uncon-strained mobility will decrease the positioning accuracy ofPDR

94 Power Consumption The power consumption of thecrowd sensing approach consists of two parts sensing con-sumption and localization consumption In order to gen-erate a dense fingerprint database high rate of samplingis demanded however which will fast drain the batteryOn the other part high frequency location estimation cankeep the trajectory smooth and continuous but consumemore power The trade-off between power consumption andsamplinglocalization rate should be investigated

95 Low Cost Sensors Most built-in sensors in the smart-phone are of low cost The performance of consumer sen-sors is surely lower than those of specified sensors Inorder to achieve a satisfactory positioning performance therequirement of algorithms is higher than that of professionalsensors and the additional information should be integratedto improve the performance

96 High-Dimensional Data The dimension of crowd sens-ing data is dominated by three elements the number ofparticipants data volume of a participant continuously con-tributing and the size of features extracted from varyingopportunistic signals used for fingerprint database genera-tion If a large number of participants continuously con-tribute multisources data with a high sampling rate thismight increase the risk of dimension disaster Incrementallearning algorithms and feature selection methods should befurther researched to keep data dimension at a controllablelevel

97 Participation Willingness High participation willingnesswill bring massive contributions However users do not havethe enthusiasm to participate because of the privacy issuepower consumption problem and so onTherefore solutionssuch as game-based coupon reward and earning credits areutilized to encourage the data contribution

98 Privacy Protection As discussed above the privacy issueis one of the factors which hold the users back for datacontributionThe data such as locations and motion patternsof a participant can be further used for inferring the sensitivepersonal information for instance habits hobbies healthyand so on Therefore privacy protection must be seriouslytreated in the crowd sensing approach

10 Conclusion and Future Trends

This survey discusses the crowd sensing based mobile indoorlocalization in terms of foundational knowledge signalsof fingerprints trajectory of obtaining fingerprints indoormaps evolution of a fingerprint database positioning algo-rithms state-of-the-art solutions and challenges In lastyears increasing researchers start to pay their attention to thecrowd sensing based indoor localization relevant topics Eventhough the crowd sensing concept is widely accepted thereare a lot of unsolved problems to transfer the concept into apractical system

Nowadays differential methods and some calibrationmethods are studied or applied for solving the problem ofdevices diversity which improve the stability of the finger-prints on the condition of losing some information of rawmeasurements In order to achieve an accurate trajectoryof a participant using a smartphone without inventions thenatural PDR which is a pedestrian dead reckoning methodthat can be applied during user living activities less or withoutconstraint will be further studied in the future Natural PDRoutputs and increasing signals will be combined with SLAMalgorithms to obtain the signalmap anduser trajectory simul-taneously Obviously data fusion is the most challenging taskwith increasing volume of the crowd Data quality controland fusion algorithms are facing lack of attention currentlyA large number of signal snapshots might be contributed byparticipants who occasionally use anAPPwith crowd sensingcapability in a short time Using the sparse and contextlesssignal snapshots to maintain an organic fingerprint databaseis a problem missed by researchers In general researcherswill pay attention to data fusion of big spatial data and signal


features natural trajectory obtaining and multiple signalscombination in the future

Competing Interests

The authors declare that they have no competing interests

Acknowledgments

This work is supported by the National Natural ScienceFoundation of China under Grants 61573242 and 61402283and in part by the Shanghai Science and Technology Com-mittee under Grants 14511100300 and 15511105100 and partlysponsored by Shanghai Pujiang Program (no 14PJ1405000)

References

[1] L Pei R Chen J Liu et al ldquoMotion recognition assisted indoorwireless navigation on a mobile phonerdquo in Proceedings of the23rd International Technical Meeting of the Satellite Division ofthe Institute of Navigation pp 3366ndash3375 Portland Ore USASeptember 2010

[2] J Liu R Chen Y Chen L Pei and L Chen ldquoiParking anintelligent indoor location-based smartphone parking servicerdquoSensors vol 12 no 11 pp 14612ndash14629 2012

[3] L Pei J Liu R Guinness Y Chen H Kuusniemi and R ChenldquoUsing LS-SVM based motion recognition for smartphoneindoor wireless positioningrdquo Sensors vol 12 no 5 pp 6155ndash6175 2012

[4] L Ruotsalainen H Kuusniemi and R Chen ldquoVisual-aidedtwo-dimensional pedestrian indoor navigation with a smart-phonerdquo Journal of Global Positioning Systems vol 10 pp 11ndash182011

[5] A Mulloni D Wagner I Barakonyi and D SchmalstiegldquoIndoor positioning and navigation with camera phonesrdquo IEEEPervasive Computing vol 8 no 2 pp 22ndash31 2009

[6] H Zhou D Zou L Pei R Ying P Liu and W Yu ldquoStruct-SLAM visual SLAMwith building structure linesrdquo IEEE Trans-actions on Vehicular Technology vol 64 no 4 pp 1364ndash13752015

[7] IndoorAtlas Ltd Oulu Finland December 2015 httpwwwindooratlascom

[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 (INFOCOM rsquo00) pp 775ndash784March2000

[9] K Pahlavan F Akgul Y Ye et al ldquoTaking positioning indoorsWi-Fi localization andGNSSrdquo Inside GNSS vol 5 no 3 pp 40ndash47 2010

[10] Ekahau Inc httpwwwekahaucom[11] L Pei R Chen J Liu H Kuusniemi T Tenhunen and Y Chen

ldquoUsing inquiry-based Bluetooth RSSI probability distributionsfor indoor positioningrdquo Journal of Global Positioning Systemsvol 9 no 2 pp 122ndash130 2010

[12] B N Schilit A LaMarca G Borriello et al ldquoChallenge ubiqui-tous location-aware computing and the lsquoplace labrsquo initiativerdquo inProceedings of the 1st ACM International Workshop on WirelessMobile Applications and Services on WLAN Hotspots (WMASHrsquo03) P Kermani Ed pp 29ndash35 ACM San Diego Calif USA2003

[13] L Von Ahn B Maurer C McMillen D Abraham and MBlum ldquoreCAPTCHA human-based character recognition viaweb security measuresrdquo Science vol 321 no 5895 pp 1465ndash1468 2008

[14] S S Kanhere ldquoParticipatory sensing crowdsourcing data frommobile smartphones in urban spacesrdquo inDistributed Computingand Internet Technology 9th International Conference ICDCIT2013 Bhubaneswar India February 5ndash8 2013 Proceedings vol7753 of Lecture Notes in Computer Science pp 19ndash26 SpringerBerlin Germany 2013

[15] A J Quinn and B B Bederson ldquoHuman computation asurvey and taxonomy of a growing fieldrdquo in Proceedings of the29th Annual CHI Conference on Human Factors in ComputingSystems (CHI rsquo11) pp 1403ndash1412 Vancouver Canada May 2011

[16] A Kapadia D Kotz and N Triandopoulos ldquoOpportunisticsensing security challenges for the new paradigmrdquo in Pro-ceedings of the 1st International Conference on CommunicationSystems and Networks and Workshops (COMSNETS rsquo09) pp 1ndash10 IEEE Bangalore India January 2009

[17] D GMurray E Yoneki J Crowcroft and SHand ldquoThe case forcrowd computingrdquo in Proceedings of the 2nd ACM SIGCOMMWorkshop on Networking Systems and Applications on MobileHandhelds (SIGCOMM rsquo10) pp 39ndash44 ACM August 2010

[18] A Madan M Cebrian D Lazer and A Pentland ldquoSocialsensing for epidemiological behavior changerdquo in Proceedingsof the 12th International Conference on Ubiquitous Computing(UbiComp rsquo10) pp 291ndash300 ACM Copenhagen DenmarkSeptember 2010

[19] M-R Ra B Liu T F La Porta and R Govindan ldquoMedusaa programming framework for crowd-sensing applicationsrdquoin Proceedings of the 10th International Conference on MobileSystems Applications and Services (MobiSys rsquo12) pp 337ndash350Lake District United Kingdom June 2012

[20] J Liu R Chen L Pei R Guinness and H Kuusniemi ldquoAhybrid smartphone indoor positioning solution for mobileLBSrdquo Sensors vol 12 no 12 pp 17208ndash17233 2012

[21] Y Yuan L Pei C Xu Q Liu and T Gu ldquoEfficient WiFi finger-print training using semi-supervised learningrdquo in Proceedingsof the Ubiquitous Positioning Indoor Navigation and LocationBased Service (UPINLBS rsquo14) pp 148ndash155 Corpus Christ TexUSA November 2014

[22] M A Youssef A Agrawala and A Udaya Shankar ldquoWLANlocation determination via clustering and probability distribu-tionsrdquo in Proceedings of the 1st IEEE International Conferenceon Pervasive Computing and Communications (PerCom rsquo03) pp143ndash150 IEEE Fort Worth Tex USA March 2003

[23] T Roos P Myllymaki H Tirri P Misikangas and J SievanenldquoA probabilistic approach to WLAN user location estimationrdquoInternational Journal of Wireless Information Networks vol 9no 3 pp 155ndash164 2002

[24] Z Xiang S Song J Chen H Wang J Huang and X GaoldquoA wireless LAN-based indoor positioning technologyrdquo IBMJournal of Research and Development vol 48 no 5-6 pp 617ndash626 2004

[25] P Bahl and V N Padmanabhan ldquoRADAR an in-building RF-based user location and tracking systemrdquo in Proceeding of the19th IEEE Annual Joint Conference of the IEEE Computer andCommunications Societies vol 2 pp 775ndash784 2000

[26] J Liu Y Chen A Jaakkola et al ldquoThe uses of ambient lightfor ubiquitous positioningrdquo in Proceedings of the IEEEIONPosition Location and Navigation Symposium (PLANS rsquo14) pp102ndash108 IEEE Monterey Calif USA May 2014


[27] MAzizyan I Constandache andR R Choudhury ldquoSurround-Sense mobile phone localization via ambience fingerprintingrdquoin Proceedings of the 15th Annual ACM International Conferenceon Mobile Computing and Networking (MobiCom rsquo09) pp 261ndash272 Beijing China September 2009

[28] J Qian L Pei J Ma R Ying and P Liu ldquoVector graphassisted pedestrian dead reckoning using an unconstrainedsmartphonerdquo Sensors vol 15 no 3 pp 5032ndash5057 2015

[29] B Ferris D Fox and N D Lawrence ldquoWiFi-SLAM usinggaussian process latent variable modelsrdquo in Proceedings ofthe 20th International Joint Conference on Artifical Intelligence(IJCAI rsquo07) vol 7 pp 2480ndash2485 January 2007

[30] P Robertson M Frassl M Angermann et al ldquoSimultaneouslocalization and mapping for pedestrians using distortions ofthe local magnetic field intensity in large indoor environ-mentsrdquo in Proceedings of the International Conference on IndoorPositioning and Indoor Navigation (IPIN rsquo13) pp 1ndash10 IEEEMontbeliard France October 2013

[31] M Montemerlo S Thrun D Koller and B Wegbreit ldquoFast-SLAM a factored solution to the simultaneous localization andmapping problemrdquo in Proceedings of the 18th National Confer-ence on Artificial Intelligence (AAAI rsquo02) and the 14th InnovativeApplications of Artificial Intelligence Conference on ArtificialIntelligence (IAAI rsquo02) pp 593ndash598 Edmonton Canada July-August 2002

[32] G Grisetti R Kummerle C Stachniss and W Burgard ldquoAtutorial on graph-based SLAMrdquo IEEE Intelligent TransportationSystems Magazine vol 2 no 4 pp 31ndash43 2010

[33] R M Faragher and R K Harle ldquoTowards an efficient intel-ligent opportunistic smartphone indoor positioning systemrdquoNavigation vol 62 no 1 pp 55ndash72 2015

[34] J-G Park B Charrow D Curtis et al ldquoGrowing an organicindoor location systemrdquo in Proceedings of the 8th AnnualInternational Conference on Mobile Systems Applications andServices (MobiSys rsquo10) pp 271ndash284 ACM June 2010

[35] G Chatzimilioudis A Konstantinidis C Laoudias and DZeinalipour-Yazti ldquoCrowdsourcing with smartphonesrdquo IEEEInternet Computing vol 16 no 5 pp 36ndash44 2012

[36] T Gallagher B Li A G Dempster and C Rizos ldquoDatabaseupdating through user feedback in fingerprint-based Wi-Filocation systemsrdquo in Proceedings of the Ubiquitous PositioningIndoor Navigation and Location Based Service (UPINLBS rsquo10)pp 1ndash8 IEEE Kirkkonummi Finland October 2010

[37] Y Kim Y Chon and H Cha ldquoSmartphone-based collaborativeand autonomous radio fingerprintingrdquo IEEE Transactions onSystems Man and Cybernetics Part C Applications and Reviewsvol 42 no 1 pp 112ndash122 2012

[38] M Allahbakhsh B Benatallah A Ignjatovic H R Motahari-Nezhad E Bertino and S Dustdar ldquoQuality control in crowd-sourcing systems issues and directionsrdquo IEEE Internet Comput-ing vol 17 no 2 pp 76ndash81 2013

[39] V C Raykar S Yu L H Zhao et al ldquoLearning from crowdsrdquoThe Journal of Machine Learning Research vol 11 pp 1297ndash13222010

[40] Y Bachrach T Graepel G Kasneci M Kosinski and JVan Gael ldquoCrowd IQ aggregating opinions to boost perfor-mancerdquo in Proceedings of the 11th International Conference onAutonomous Agents and Multiagent SystemsmdashVolume 1 pp535ndash542 International Foundation forAutonomousAgents andMultiagent Systems Valencia Spain June 2012

[41] E Kamar S Hacker and E Horvitz ldquoCombining humanand machine intelligence in large-scale crowdsourcingrdquo in

Proceedings of the 11th International Conference on AutonomousAgents and Multiagent Systems (AAMAS rsquo12) vol 1 pp 467ndash474 International Foundation for Autonomous Agents andMultiagent Systems 2012

[42] P Welinder S Branson P Perona and S J Belongie ldquoThemultidimensional wisdom of crowdsrdquo in Advances in NeuralInformation Processing Systems pp 2424ndash2432MITPress 2010

[43] J Whitehill T F Wu J Bergsma J R Movellan and P LRuvolo ldquoWhose vote should count more optimal integrationof labels from labelers of unknown expertiserdquo in Advances inNeural Information Processing Systems pp 2035ndash2043 2009

[44] S J Julier and J K Uhlmann ldquoGeneral decentralized datafusion with covariance intersection (CI)rdquo in Handbook of DataFusion CRC Press Boca Raton Fla USA 2001

[45] J K Uhlmann ldquoCovariance consistency methods for fault-tolerant distributed data fusionrdquo Information Fusion vol 4 no3 pp 201ndash215 2003

[46] G Shakhnarovich T Darrell and P Indyk ldquoNearest-neighbormethods in learning and visionrdquo IEEE Transactions on NeuralNetworks vol 19 no 2 p 377 2008

[47] Y Kou C T Lu and D Chen ldquoSpatial weighted outlierdetectionrdquo in Proceedings of the SIAM International Conferenceon Data Mining (SDM rsquo06) pp 614ndash618 April 2006

[48] M M Breunig H-P Kriegel R T Ng and J Sander ldquoLOFidentifying density-based local outliersrdquo ACM Sigmod Recordvol 29 no 2 pp 93ndash104 2000

[49] M Venanzi A Rogers and N R Jennings ldquoTrust-based fusionof untrustworthy information in crowdsourcing applicationsrdquoin Proceedings of the International Conference on AutonomousAgents and Multi-agent Systems (AAMAS rsquo13) pp 829ndash836Saint Paul MN USA May 2013

[50] K-C Chou and C-T Zhang ldquoPrediction of protein structuralclassesrdquo Critical Reviews in Biochemistry andMolecular Biologyvol 30 no 4 pp 275ndash349 1995

[51] R Chen L Pei J Liu and H Leppakoski ldquoWLAN and blue-tooth positioning in smart phonesrdquo in Ubiquitous PositioningandMobile Location-Based Services in Smart Phones pp 44ndash68IGI Global 2012

[52] J Liu R Chen L Pei et al ldquoAccelerometer assisted wirelesssignals robust positioning based on hidden markov modelrdquo inProceedings of the IEEEION Position Location and NavigationSymposium (PLANS rsquo10) pp 488ndash497 IndianWells Calif USAMay 2010

[53] H Kuusniemi J Liu L Pei Y Chen L Chen and R ChenldquoReliability considerations of multi-sensor multi-networkpedestrian navigationrdquo IET Radar Sonar and Navigation vol6 no 3 pp 157ndash164 2012

[54] M S Arulampalam S Maskell N Gordon and T Clapp ldquoAtutorial on particle filters for online nonlinearnon-GaussianBayesian trackingrdquo IEEE Transactions on Signal Processing vol50 no 2 pp 174ndash188 2002

[55] P Bolliger ldquoRedpin-adaptive zero-configuration indoor local-ization through user collaborationrdquo in Proceedings of the 1stACM International Workshop onMobile Entity Localization andTracking in Gps-Less Environments pp 55ndash60 San FranciscoCalif USA September 2008

[56] A Rai K K Chintalapudi V N Padmanabhan and R SenldquoZee zero-effort crowdsourcing for indoor localizationrdquo inProceedings of the 18th Annual International Conference onMobile Computing andNetworking (MobiCom rsquo12) pp 293ndash304ACM Istanbul Turkey August 2012


[57] Z Yang C Wu and Y Liu ldquoLocating in fingerprint spacewireless indoor localization with little human interventionrdquoin Proceedings of the 18th Annual International Conference onMobile Computing andNetworking (MobiCom rsquo12) pp 269ndash280Instanbul Turkey August 2012

[58] P Robertson M G Puyol and M Angermann ldquoCollaborativepedestrian mapping of buildings using inertial sensors andFootSLAMrdquo in Proceedings of the 24th International TechnicalMeeting of the Satellite Division of the Institute of Navigation(ION GNSS rsquo11) pp 1366ndash1377 September 2011

[59] V Radu and M K Marina ldquoHiMLoc indoor smartphonelocalization via activity aware pedestrian dead reckoning withselective crowdsourced WiFi fingerprintingrdquo in Proceedings ofthe International Conference on Indoor Positioning and IndoorNavigation (IPIN rsquo13) pp 1ndash10 IEEE Montbeliard-BelfortFrance October 2013

[60] H Wang S Sen A Elgohary M Farid M Youssef and RR Choudhury ldquoNo need to war-drive unsupervised indoorlocalizationrdquo in Proceedings of the 10th International Conferenceon Mobile Systems Applications and Services (MobiSys rsquo12) pp197ndash210 ACM June 2012

[61] R M Faragher and R K Harle ldquoSmartSLAMmdashan efficientsmartphone indoor positioning system exploiting machinelearning and opportunistic sensingrdquo in Proceedings of the26th International Technical Meeting of the Satellite Division ofthe Institute of Navigation (ION GNSS rsquo13) vol 13 pp 1ndash14September 2013

[62] S Yang P Dessai M Verma and M Gerla ldquoFreeloccalibration-free crowdsourced indoor localizationrdquo in Proceed-ings of the IEEE Conference on Computer Communications(INFOCOM rsquo13) pp 2481ndash2489 IEEE Turin Italy April 2013

[63] M Lee S H Jung S Lee and D Han ldquoElekspot a platformfor urban place recognition via crowdsourcingrdquo in Proceedingsof the IEEEIPSJ 12th International Symposium on Applicationsand the Internet (SAINT rsquo12) pp 190ndash195 Izmir Turkey July2012

[64] J Niu BWang L Cheng et al ldquoWicLoc an indoor localizationsystem based on WiFi fingerprints and crowdsourcingrdquo inProceedings of the IEEE International Conference on Communi-cations (ICC rsquo15) pp 3008ndash3013 London UK June 2015

[65] J Machaj P Brida and R Piche ldquoRank based fingerprintingalgorithm for indoor positioningrdquo in Proceedings of the Interna-tional Conference on Indoor Positioning and Indoor Navigation(IPIN rsquo11) pp 1ndash6 Guimaraes Portugal September 2011

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



Observable signals

Location

Fingerprint

Current measurements

Fingerprint DB

Feature matching

Pos

Learning phase Positioning phase

Observable vector

Sam

ples Feature vectors

Loca

tions∙ Extracted features

∙ Location

middot middot middotmiddot middot middot




Figure 1 Flowchart of fingerprinting-based localization approach

pervasive providing the existing infrastructures RF-basedopportunistic signals have been widely applied for indoorlocalization for indoor environments [12]

(2) Crowd Sensing In 2005 with outsources fast growingcrowdsourcing has emerged as a new collaborative paradigmin which crowds of people can collaborate and accomplisha specific task The most famous crowdsourcing projectis the reCAPTCHA (Completely Automated Public Turingtest to Tell Computers and Humans Apart) which is themost applied technology of preventing malicious codes fromabusing web services [13] The project achieves such goal byrequesting the user input such as the distorted characterswhich can only be performed by human being so far Themost elegant design of the reCAPTCHA is to transformsuch human load into a useful application where the userinputs can be utilized to digitize old publications which arehardly recognized by optical character recognition (OCR)Normally userrsquos inputs are the distorted fragment scannedfrom those old publications The reCAPTCHA toolkit hadbeen deployed in more than 40000 websites With theinvolvements of crowd web users the toolkit had digitalizedover 440 million words by 2008 In results the archivesof New York Times over 150 years were unconsciouslyrecognized and digitalized by web users within a fewmonths

Skyhook a worldwide Wi-Fi positioning provider col-lects the access points and their locations contributed byparticipants via crowdsourcing [9] The contribution will beused to update the Wi-Fi database after the quality controlcenter verifies the crowdsourced information Similar tocrowdsourcing the terms such as participatory sensing [14]human computation [15] opportunistic sensing [16] crowdcomputing [17] social sensing [18] and crowd sensing [19]have a common idea which is to aggregate the collectiveintelligence of a crowd to achieve a goal with inexpensive costUnconscious participation is the profile which fundamentallydistinguishes crowd sensing from other crowdsourcing-like

solutions which makes crowd sensing an ideal approach togenerating fingerprints for indoor localization by amountof common mobile phone users instead of expert sitesurveyors

(3) Organization of This Paper In this survey we review thestate-of-the-art indoor localization research related to crowdsensing solutions The contents are structured as followsSection 2 introduces the foundation of crowd sensing indoorlocalization in high level Section 3 examines the possiblesignals for fingerprints of indoor localization Section 4surveys the methods for obtaining the walking trajectory ofa participant Section 5 discusses the types of map used forindoor localization Section 6 looks at how the fingerprintsorganically change thewidely applied positioning algorithmsare discussed in Section 7 Section 8 compares the state-of-the-art solutions published recent years Section 9 points outthe challenges of crowd sensing based indoor localizationand then Section 10 concludes by identifying open researchtopics and future research directions

2 Foundation of Crowd Sensing forIndoor Localization

Crowd sensing is a paradigm using crowd contributionto achieve a complex task which is perfectly suitable forfingerprinting-based indoor localization To generate a timelyfingerprint database also known as a radio map in theWi-Fi localization this needs regular site surveys with anexpert in the target area In the crowd sensing solution theprofessional site survey phase is supposed to be replaced withnonprofessional sensing using off-shelf mobile devices suchas smartphone

Figure 1 describes a typical flowchart of fingerprinting-based localization approach which consists of two phasesnamely learning phase and positioning phase respectivelyDuring the learning phase also known as training phase


FC1 FCn

FCi

IndoormapFP DB


Positionfeatures






OS1(eg Wi-Fi)

OS2 (egBluetooth)



OSn (eg cellular)















































119909119896+1 = 119909119896 + SL119896 sin 120579119896

119910119896+1 = 119910119896 + SL119896 cos 120579119896(1)





119901 (X1119905M | D1119905Z1119905 1199090) (2)


5 Indoor Maps



















R119894 =

11987411 sdot sdot sdot 1198741119896

d

1198741198991 sdot sdot sdot 119874119899119896

(3)


R119894 = []1198941 ]119894119902] (4)



F =

[

[

[

[

[

X1 R1

X119898 R119898

]

]

]

]

]

(5)


119889 (R119888 R119894) =

10038171003817100381710038171003817

R119888 minus R11989410038171003817100381710038171003817 (6)



119897 (R119888) = argmin

R119894isinP119889 (

R119888 R119894) (7)


119909 (R119888) =

1

119896

119896

sum

119894=1

119897119894 (R119888) (8)



R119894

=

[

[

[

[

[

[

[


d


]

]

]

]

]

]

]

(9)



F = [R1R2 R119908] (10)




WLAN

SensorsCamera

Accelerometer

Gyroscope


DRPDR techniques


Calibration

(S O A B 120587)


Output


Knowledgedatabase

GNSS RFID




Initializing

Integrity







119875 (O | X) 119875 (X)119875 (O)

] (11)






119875 (O | X) =119896

prod

119899=1

119875 (119878119899 119874119899 | X) (14)









Sign

al st

reng

th (d

B)

minus50


minus60

87 6 5 4 3 2 1 0

876543210

Y (m)X (m)

Locations


GPS

Mag

Wi-Fi

Gyro


Particle Ini

Particle Upt

ResamplingError

est


Stepdetection


Headingestimation

Locationest

PDR

+Y

minusY

minusZ

+Z

minusX

+X

RawData

Motion recognition








059meters















FreeLoc W SP DAlt2m

(Hallway)lt4m (Lab)










9 Challenges















Competing Interests


Acknowledgments


References












































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




FC1 FCn

FCi

IndoormapFP DB


Positionfeatures






OS1(eg Wi-Fi)

OS2 (egBluetooth)



OSn (eg cellular)















































119909119896+1 = 119909119896 + SL119896 sin 120579119896

119910119896+1 = 119910119896 + SL119896 cos 120579119896(1)





119901 (X1119905M | D1119905Z1119905 1199090) (2)


5 Indoor Maps



















R119894 =

11987411 sdot sdot sdot 1198741119896

d

1198741198991 sdot sdot sdot 119874119899119896

(3)


R119894 = []1198941 ]119894119902] (4)



F =

[

[

[

[

[

X1 R1

X119898 R119898

]

]

]

]

]

(5)


119889 (R119888 R119894) =

10038171003817100381710038171003817

R119888 minus R11989410038171003817100381710038171003817 (6)



119897 (R119888) = argmin

R119894isinP119889 (

R119888 R119894) (7)


119909 (R119888) =

1

119896

119896

sum

119894=1

119897119894 (R119888) (8)



R119894

=

[

[

[

[

[

[

[


d


]

]

]

]

]

]

]

(9)



F = [R1R2 R119908] (10)




WLAN

SensorsCamera

Accelerometer

Gyroscope


DRPDR techniques


Calibration

(S O A B 120587)


Output


Knowledgedatabase

GNSS RFID




Initializing

Integrity







119875 (O | X) 119875 (X)119875 (O)

] (11)






119875 (O | X) =119896

prod

119899=1

119875 (119878119899 119874119899 | X) (14)









Sign

al st

reng

th (d

B)

minus50


minus60

87 6 5 4 3 2 1 0

876543210

Y (m)X (m)

Locations


GPS

Mag

Wi-Fi

Gyro


Particle Ini

Particle Upt

ResamplingError

est


Stepdetection


Headingestimation

Locationest

PDR

+Y

minusY

minusZ

+Z

minusX

+X

RawData

Motion recognition








059meters















FreeLoc W SP DAlt2m

(Hallway)lt4m (Lab)










9 Challenges















Competing Interests


Acknowledgments


References












































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in









































119909119896+1 = 119909119896 + SL119896 sin 120579119896

119910119896+1 = 119910119896 + SL119896 cos 120579119896(1)





119901 (X1119905M | D1119905Z1119905 1199090) (2)


5 Indoor Maps



















R119894 =

11987411 sdot sdot sdot 1198741119896

d

1198741198991 sdot sdot sdot 119874119899119896

(3)


R119894 = []1198941 ]119894119902] (4)



F =

[

[

[

[

[

X1 R1

X119898 R119898

]

]

]

]

]

(5)


119889 (R119888 R119894) =

10038171003817100381710038171003817

R119888 minus R11989410038171003817100381710038171003817 (6)



119897 (R119888) = argmin

R119894isinP119889 (

R119888 R119894) (7)


119909 (R119888) =

1

119896

119896

sum

119894=1

119897119894 (R119888) (8)



R119894

=

[

[

[

[

[

[

[


d


]

]

]

]

]

]

]

(9)



F = [R1R2 R119908] (10)




WLAN

SensorsCamera

Accelerometer

Gyroscope


DRPDR techniques


Calibration

(S O A B 120587)


Output


Knowledgedatabase

GNSS RFID




Initializing

Integrity







119875 (O | X) 119875 (X)119875 (O)

] (11)






119875 (O | X) =119896

prod

119899=1

119875 (119878119899 119874119899 | X) (14)









Sign

al st

reng

th (d

B)

minus50


minus60

87 6 5 4 3 2 1 0

876543210

Y (m)X (m)

Locations


GPS

Mag

Wi-Fi

Gyro


Particle Ini

Particle Upt

ResamplingError

est


Stepdetection


Headingestimation

Locationest

PDR

+Y

minusY

minusZ

+Z

minusX

+X

RawData

Motion recognition








059meters















FreeLoc W SP DAlt2m

(Hallway)lt4m (Lab)










9 Challenges















Competing Interests


Acknowledgments


References












































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



















R119894 =

11987411 sdot sdot sdot 1198741119896

d

1198741198991 sdot sdot sdot 119874119899119896

(3)


R119894 = []1198941 ]119894119902] (4)



F =

[

[

[

[

[

X1 R1

X119898 R119898

]

]

]

]

]

(5)


119889 (R119888 R119894) =

10038171003817100381710038171003817

R119888 minus R11989410038171003817100381710038171003817 (6)



119897 (R119888) = argmin

R119894isinP119889 (

R119888 R119894) (7)


119909 (R119888) =

1

119896

119896

sum

119894=1

119897119894 (R119888) (8)



R119894

=

[

[

[

[

[

[

[


d


]

]

]

]

]

]

]

(9)



F = [R1R2 R119908] (10)




WLAN

SensorsCamera

Accelerometer

Gyroscope


DRPDR techniques


Calibration

(S O A B 120587)


Output


Knowledgedatabase

GNSS RFID




Initializing

Integrity







119875 (O | X) 119875 (X)119875 (O)

] (11)






119875 (O | X) =119896

prod

119899=1

119875 (119878119899 119874119899 | X) (14)









Sign

al st

reng

th (d

B)

minus50


minus60

87 6 5 4 3 2 1 0

876543210

Y (m)X (m)

Locations


GPS

Mag

Wi-Fi

Gyro


Particle Ini

Particle Upt

ResamplingError

est


Stepdetection


Headingestimation

Locationest

PDR

+Y

minusY

minusZ

+Z

minusX

+X

RawData

Motion recognition








059meters















FreeLoc W SP DAlt2m

(Hallway)lt4m (Lab)










9 Challenges















Competing Interests


Acknowledgments


References












































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





WLAN

SensorsCamera

Accelerometer

Gyroscope


DRPDR techniques


Calibration

(S O A B 120587)


Output


Knowledgedatabase

GNSS RFID




Initializing

Integrity







119875 (O | X) 119875 (X)119875 (O)

] (11)






119875 (O | X) =119896

prod

119899=1

119875 (119878119899 119874119899 | X) (14)









Sign

al st

reng

th (d

B)

minus50


minus60

87 6 5 4 3 2 1 0

876543210

Y (m)X (m)

Locations


GPS

Mag

Wi-Fi

Gyro


Particle Ini

Particle Upt

ResamplingError

est


Stepdetection


Headingestimation

Locationest

PDR

+Y

minusY

minusZ

+Z

minusX

+X

RawData

Motion recognition








059meters















FreeLoc W SP DAlt2m

(Hallway)lt4m (Lab)










9 Challenges















Competing Interests


Acknowledgments


References












































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Sign

al st

reng

th (d

B)

minus50


minus60

87 6 5 4 3 2 1 0

876543210

Y (m)X (m)

Locations


GPS

Mag

Wi-Fi

Gyro


Particle Ini

Particle Upt

ResamplingError

est


Stepdetection


Headingestimation

Locationest

PDR

+Y

minusY

minusZ

+Z

minusX

+X

RawData

Motion recognition








059meters















FreeLoc W SP DAlt2m

(Hallway)lt4m (Lab)










9 Challenges















Competing Interests


Acknowledgments


References












































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in
















FreeLoc W SP DAlt2m

(Hallway)lt4m (Lab)










9 Challenges















Competing Interests


Acknowledgments


References












































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




9 Challenges















Competing Interests


Acknowledgments


References












































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





Competing Interests


Acknowledgments


References












































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




















































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



review article a survey of crowd sensing opportunistic...

Documents