Journal of Compuer and Communicaions, 207, 5, 02-5 hp://www.scirp.org/journal/jcc ISSN Online: 2327-5227 ISSN Prin: 2327-529 An Indoor Pedesrian Localizaion Algorihm Based on Muli-Sensor Informaion Fusion Xiangyu Xu,2, Mei Wang,2, Liyan Luo,2, Zhibin Meng,2, Enliang Wang,2 Key Laboraory of Cogniive Radio & Informaion Processing, Minisry of Educaion, Guilin Universiy of Elecronic Technology, Guilin, China 2 Guangxi Experimen Cener of Informaion Science, Guilin, China How o cie his paper: Xu, X.Y., Wang, M., Luo, L.Y., Meng, Z.B. and Wang, E.L. (207) An Indoor Pedesrian Localizaion Algorihm Based on Muli-Sensor Informaion Fusion. Journal of Compuer and Communicaions, 5, 02-5. hps://doi.org/0.4236/jcc.207.5302 Received: February 3, 207 Acceped: March 0, 207 Published: March 3, 207 Absrac For exising indoor localizaion algorihm has low accuracy, high cos in deploymen and mainenance, lack of robusness, and low sensor uilizaion, his paper proposes a paricle filer algorihm based on muli-sensor fusion. The pedesrian s localizaion in indoor environmen is described as dynamic sysem sae esimaion problem. The algorihm combines he smar mobile erminal wih indoor localizaion, and filers he resul of localizaion wih he paricle filer. In his paper, a dynamic inerval paricle filer algorihm based on pedesrian dead reckoning (PDR) informaion and RSSI localizaion informaion have been used o improve he filering precision and he sabiliy. Moreover, he localizaion resuls will be uploaded o he server in ime, and he locaion fingerprin daabase will be buil incremenally, which can adap he dynamic changes of he indoor environmen. Experimenal resuls show ha he algorihm based on muli-sensor improves he localizaion accuracy and robusness compared wih he locaion algorihm based on Wi-Fi. Keywords Muli-Sensor Fusion, Indoor Localizaion, Pedesrian Dead Reckoning (PDR), Paricle Filer. Inroducion The indoor posiioning navigaion sysem can provide navigaion service for users in public places such as large complex buildings, and has wide applicaion prospec [] [2]. There has been a growing ineres in indoor posiioning echnology ha relies on he exising senor, like he Wi-Fi, Zigbee, Pedesrian dead reckoning (PDR), Received signal srengh indicaion (RSSI) and Radio Frequency Idenificaion (RFID). As he PDR posiioning sysems can only provide DOI: 0.4236/jcc.207.5302 March 3, 207
relaive posiion informaion, error will accumulae over ime, i is necessary o provide absolue posiion informaion o correc he error [3]. RSSI posiioning algorihm is simple, can provide absolue locaion informaion wihou adding addiional hardware, Therefore, he fusion algorihm based on PDR and RSSI has been widespread concern. Paper [4] discussed Indoor Locaion Algorihm based on he RSSI fingerprin informaion, and his algorihm has high accuracy only in low noise environmen, i is no suiable for high noise environmen. Paper [5] analyzes he influence of pah aenuaion coefficien on locaion accuracy in order o improve he accuracy in he high noise environmen. In paper [6], we propose a mehod o calculae he pah fading exponen by measuring he node energy and he geomeric relaionship among nodes. In recen decades, wih he rapid developmen of inegraed circuis, smar phones had made grea progress in daa sorage and daa processing, and embedded many micro-sensors, such as he acceleromeer, he gyroscopes, and he magneomeers and so on [9]. The rapid developmen of smarphones provides a new plaform and opporuniy o achieve an economical and friendly posiioning sysem in oday s indoor environmen. This paper mainly sudies a paricle filer algorihm based on muli-sensor fusion indoor pedesrian localizaion, and combines he smarphone wih he radiional posiioning echnology. The firs sep, he sensors buil-in smarphone can predic he user s movemen and observaion saus, as Bayesian esimaes of he movemen model and observaion model, and esablish he fingerprin daabase of indoor environmen. The second sep, he paricle filer algorihm can filer and fusion he movemen model and he observaion model. The las sep, he localizaion resuls will be uploaded o he server in ime, and he locaion fingerprin daabase will be buil incremenally, which can adap he dynamic ransform of he indoor environmen. 2. Relaed Work 2.. Indoor Locaion Algorihm Based on Muli-Sensor Informaion The sensors buil-in smarphone, such as acceleraion sensors, gyroscopes, graviaional acceleraion and magneomeers, can rack he locaion of he pedesrian in he indoor environmen. Figure shows he overall sysem block diagram. The wireless module and oher sensors buil-in he smarphone can predic he pedesrian s environmen sae and he pedesrian s movemen sae. The proposed algorihm can fuse informaion of he pedesrian s movemen and observaion, and upload he pedesrian s locaion informaion o he applicaion layer. 2.2. Wi-Fi Fingerprin Locaion Algorihm This algorihm is usually divided ino wo sages: he offline raining phase and online posiioning phase. In he firs sage, we should se many reference poins 03
Applicaion layer LBS applicaion ohers Inermediae layer Sep deecion Sep calibraion direcion PDR Paricle filer Fingerprin Locaion Sensor Acceleromeer Magneomeer Magneomeer Wi-Fi Figure. Algorihm archiecure. in his sage and collec he reference daa from Wi-Fi access poin (AP), such as signal srengh, arrival angle, and frequency and so on. Nex, we sore he reference daa wih he locaion informaion ino he daabase as a se of fingerprin daa. In he second sage, we use he smarphone o deec he signal daa received a he locaion o be deermined, and hen compare he signal daa wih he daabase hrough he corresponding algorihm. Nex, we ge he user s acual locaion informaion. Figure 2 shows he process of Wi-Fi fingerprin orienaion. Table shows fingerprin daabase wih class labels. The auhors in [7] proposed ha he number and locaions of APs, physical layou, and mean of RSSs a RPs have significan impac on localizaion precision. To opimize he AP placemen, he auhors in [8] proposed a novel approach by using a small number of APs o provide full coverage while locaing he mobile device wihin an area wih limied size. Wang and Lin [9] proposed a goal programming-driven model which is inergraded wih a geneic algorihm and an embedded mask mechanism o resolve he problems of muliple objecive AP deploymen consrucion and enhancemen. Therefore, he error bounds analysis under differen signal disribuions in Wi-Fi environmens remains an open problem. 3. Paricle Filer Based Muli-Sensor Fusion This paper improves he posiioning precision of PDR sysem by fusion pedesrian gai informaion, indoor environmen informaion and RSSI, because of PDR sysem canno ge absolue posiion informaion. We assume he pedesrian s iniial posiion, and ge he relaive posiion informaion from PDR sysem. We improve he posiioning precision of he proposed algorihm by combine he indoor environmen o filer he posiioning resuls. Bu he above problem is a ypical nonlinear problem, his means ha he convenional linear fusion algorihm canno ge ousanding resuls. To solve hose problems, we selec he paricle filer algorihm o fuse he muli-sensor daa, which no only has 04
Training phase AP2 AP Sore DB AP3 AP4 Online Locaion Algorihm locaion Figure 2. Wi-Fi fingerprin orienaion. Table. Fingerprin daabase wih class labels. Daabase AP AP 2 AP n Class (, ) x y RSSI RSSI 2 RSSI C n (, 2 2) x y RSSI RSSI 2 22 RSSI C 2n 2 (, n n) x y RSSIn RSSI2n RSSInn Cn beer flexibiliy, exensive pracicaliy, bu also can improve he posiioning precision. 3.. Basic Mahemaical Model The pedesrian s localizaion is described as dynamic sysem sae esimaion problem [0] in he indoor environmen, he sae space equaion is defined as follows: Moion equaion: Observaion equaion: x = f ( x, w ) () z = h ( x, v ) (2) where f and h are nonlinear funcions, w and v are independen of he noise sequence, respecively he process noise and observaion noise, x is he movemen informaion a ime, z is he observaion informaion a ime. According o Bayesian esimaes, he sae esimaion problem is o deduce x from z: {,, = z z} a ime, he esimaed p (x z : ). In his paper, he paricle filer can achieve he inegraion of locaion informaion o avoid he inegral operaion in Bayesian esimaion, and provide locaion daa for indoor locaion-based services. 05
3.2. Movemen Model Due o he complexiy of he pedesrian walk and he low cos of he posiioning sysem in he indoor environmen, we use he riangulaion of sep and azimuh o obain he relaive posiion informaion of pedesrians in his paper. x+ = x + l sinθ (3) y+ = y + l cosθ where x and y represen he coordinaes in he wo-dimensional coordinae sysem, l represens he sep size a ime, θ indicaes he direcion of movemen a ime. Which mainly need o solve hree problems: sep numbers, sep lengh and moion deecion. The pedesrian s movemen behavior includes he movemen direcion and he movemen disance. The daa of he acceleraion and he angular velociy will ransform when person in he process of walking []. In his paper, we use he acceleraion sensors, gyroscopes, graviaional acceleraion and magneomeers o perceive he user s moion behavior. The acceleraion sensors are able o deermine he changes in pedesrian s acceleraion, i is represened by he acceleraion componen of he hree-dimensional direcion. The human body sep model is esablished by daa which is colleced by acceleraion sensors filering and feaure exracion. The direcion sensor can capure changes in he direcion of moion of he pedesrian. This paper uses he empirical formula o calculae he sep size, and correc he user s sep size dynamically [2]. 3.3. Observaion Model The KNN localizaion algorihm calculaes he cosine similariy of he RSSI vecor measured a he anchor poin and he RSSI vecor measured a each RP poin. The cosine similariy of he wo vecors is shown below: x y cos ( xy, ) = (4) x y And hen hrough he query fingerprin library o find he mos similar k fingerprin daa. Tha each RSSI vecor in he fingerprin library uniquely corresponds o he locaion informaion of a reference poin. The posiion of he final anchor is esimaed as he weigh of he k reference poins, he weigh of i reference poin is w i. si wi = (5) k s j= s i is cosine similariy of he i reference poin and curren poin. The posiion of he final anchor is esimaed: k x = wi xi i= (6) k y = wi yi i= j 06
3.4. Paricle Filer Based Muli-Sensor Fusion In his paper, paricle filer [3] is used o approximae he probabiliy densiy funcion of he user s posiion, and filer he movemen direcion. I is assumed ha each paricle has he following sae informaion: [,,w, ] The user s posiion of he map is (, ) x y, w is he weigh of paricle, θ is L is sep lengh, herefore, The sae ransiion equa- he movemen direcion, ion is shown below: T X = x y θ (7) i i i i x x L cosθ nx i i i i i X = y y L sinθ n = + + y i i i θ θ θ n θ (8) i is he paricle number, is gai cycle, nx, ny, n θ are Gaussian whie noise wih zero mean. 3.4.. Iniializaion Assume ha he user s iniial posiion is ( x0, y 0), he number of paricle is N, and each paricle conains he user s locaion informaion, direcion, and paricle weigh, as discussed in (7). We can obain he weigh of paricle is N, he sum of he weighs is. Each paricle represens a possible movemen sae of pedesrian, ha is, a possible locaion of pedesrian [4] [5]. 3.4.2. Paricle Sae Transiion I is he user s locaion wih he updae process over ime. In he process of pedesrian movemen, inroducing he weigh of paricle as he smooh facor of PDR sysem. As we know, differen paricle weighs will produce differen pahs. θ is he pedesrian direcion, facor. ( w ) θ = θ + w θ (9) i i i i θ i is he pedesrian direcion wih smooh 3.4.3. Paricle Updae By observing he environmen in which each paricle spreads, o verify ha he propagaion of he paricles is reasonable, and observe he degree of similariy beween he possible posiion of pedesrians represened by each paricle and he acual locaion of pedesrians. The paricles closer o he rue posiion of he pedesrian will be given a larger weigh, and vice versa. The paricle weigh is calculaed as follows: ( i i, ) i i ( x a) + ( y b) 2 2 i c w = e (0) c x y is he possible posiion of pedesrians represened by i-h paricle, ab, is he locaion which is calculaed by he fingerprin algorihm. Normalized weigh calculaion: ( ) 07
w i = w i N i w n= () 3.4.4. Posiion Calculaion Pedesrian locaion deerminaion can be based on wo crieria: Maximum poserior probabiliy and Weighed Crieria. This paper uses he second mehod o calculae he posiion. X Y = N i i = xw i= 0 N i= 0 yw i i (2) 3.4.5. Resampling Wih he increase of filering ime, he degradaion of paricles will occur, he imporance of weigh may be concenraed o a small number of paricles, he need for is resampling, increase he weigh of he larger number of paricles. 3.4.6. Algorihm Flowchar Based on he above, an indoor localizaion algorihm based on muli-sensor fusion is proposed, as shown in Figure 3. Sar Inpu Muli-sensor daa Sep deecion N Y Fingerprin daa Sep numbers, Lengh and direcion Wi-Fi fingerprin locaion Algorihm pedesrian dead reckoning (PDR) sysem Y Paricle filer Updae he Fingerprin daabase Oupu posion coniune End N Figure 3. Algorihm flowchar. 08
4. Experimenal Design and Resul Analysis In order o es he posiioning performance of he proposed algorihm in his paper, a lo of experimens are carried ou. Here are he specific experimenal conens of esing he posiioning performance based on he soluion o realize pedesrian indoor posiioning, and analyze he experimenal resuls. This paper chooses he eighh eaching building of Guilin Universiy of Elecronic Science and Technology as an experimenal sie. The experimenal sie is 2 meers long and 8 meers wide, and four wireless rouers were insalled in his experimenal sie. Experimenal sie srucure shown in Figure 4. In his experimen, he experimenal sie will be divided ino 50 * 50 small squares according o he laying of floor iles, 50 small squares in he horizonal direcion and 20 in he verical direcion. The size of each small laice is 0.5 m * 0.5 m. Daa acquisiion of acceleraion, direcion, and Wi-Fi signal srengh is achieved by programming based on Android 4.5. The experimener who hand- AP- 83 832 AP-2 AP-3 8309 830 AP-4 Figure 4. Experimenal sie srucure. 09
held Huawei mobile phone is walking a he normal pace in he experimenal sie according o he expeced rajecory, and his phone has buil-in Acceleraion sensors, gyroscopes, graviaional acceleraion and magneomeers. 4.. Sep Deecion In order o verify he effeciveness of he pedomeer, he experimener performed 50 imes experimens on 00 sep numbers. The pedomeer resuls is shown in Figure 5. Figure 5 shows he accuracy of sep numbers can reach more han 73%, and he error of sep numbers are mosly in one sep. 4.2. Build Fingerprin Daabase In he posiioning algorihm validaion process, he resul of locaion and Wireless signal srengh will be uploaded o he daabase in ime, and he locaion fingerprin daabase will be buil incremenally. The fingerprin daa able of locaion fingerprin daabase is shown in Table 2. In his paper, he resul of locaion will be couned by he localizaion algorihm based on muli-sensor informaion fusion, and he resul of locaion will be updaed he original informaion in he daabase, and add he ime samp. In his experimen, 000 ses of fingerprin daa were colleced a each observaion poin, and he colleced daa is processed by Gaussian filering in order o improve he accuracy of fingerprin daa. 4.3. Locaion Experimen The experimener is walking in he experimenal sie wih expeced rajecory, and collecing he informaion include acceleraion, direcion and Wi-Fi signal srengh. 30 00 sep es Couns 20 0 0 95 96 97 98 99 0000203 Number of seps Figure 5. Pedomeer resuls hisogram. Table 2. Fingerprin daa able of locaion fingerprin daabase. Posiion Time AP AP 2 AP 3 AP 4 (.2,.) 206--2 0:32:3 35.6667 70.68 7.4524 67.5 (2.4, 7.8) 206--2 0:32:24 38.258 68.7749 72.597 65.9253 (4.5, 5.25) 206--2 0:32:38 47.4828 68.205 75.807 64.9457 (5.0, 6.88) 206--2 0:33:05 46.8 6.3243 73 68.3333 (.2, 3.09) 206--2 0:33:7 59.092 68.7097 74.3265 75.63 0
Figure 6 and Figure 7 shows he resuls of paricle filer localizaion algorihm based on muli-sensor fusion and fingerprin posiioning algorihm, and he localizaion errors are shown in Figure 8 and Figure 9. Afer he paricles converge, he posiioning resuls are analyzed as follows. Simulaion resul by paricle filer localizaion algorihm based on muli-sensor fusion is lower in sabiliy and accuracy han he fingerprin posiioning algorihm, and he error of pedesal dead reckoning is effecively converged. The average error using muli-sensor fusion is above 0.35 meers compared o he fingerprin posiioning daa s above.65 meers. 93.9% of he localizaion errors are lower han 0.5 meer by paricle filer localizaion algorihm based on muli-sensor fusion bu he fingerprin posiioning algorihm daa s 52.3%.The average errors are shown in Table 3. Figure 6. Localizaion es rajecory Diagram (N = 00). Figure 7. Localizaion es rajecory Diagram (N = 000).
Figure 8. Error analysis of differen paricle numbers. Figure 9. Error analysis of PDR wih RSSI and RSSI. Table 3. Average error comparison of he esimaed pah and rue pah. Algorihm Average error Muli-senor fusion (N = 00) 0.35 (93.9%) Muli-senor fusion (N = 000) 0.27 (95.2%) Wi-Fi fingerprin locaion.65 (52.3%) Compared wih Figure 6 and Figure 7 we can see ha wih he increase of he number of paricles, he posiioning precision is improved furher, he average error when N = 00 is above 0.35 meers, and when N = 000 is above 0.27 meer, however, increasing he number of paricles means increasing he amoun of daa processed and he running ime of he program. When he number of paricles is 00, he program running ime is approximaely 3.78 seconds, and 2
when he number of paricles is 000, he ime increased by approximaely 5.332 seconds. This poses a grea impac on posiioning real-ime. The experimenal resuls show ha filer localizaion algorihm based on muli-sensor fusion has a higher posiioning precision han Wi-Fi fingerprin posiioning algorihm. 4.4. Robusness Verificaion In order o verify he robusness of he proposed algorihm, we also experimened o deermine he impac on posiioning accuracy wih he following: slow walking, running and he posiion of he phone. In Secion 4.3, he experimener walked a normal speed, and ook he smar phone in his hand. In he nex experimen, he experimener will walk slowly, run or pu he smar phone in pocke. Under he above hree experimenal condiions, he average errors are shown in Table 4. Table 4 shows ha he above hree cases will impac he accuracy of he posiion algorihm which proposed in his paper. Compared wih he algorihm based on muli-senor fusion and Wi-Fi we can see ha wih he increase of he number of he senor, he posiioning precision is improved furher. Bu he average errors will be increased because he sensor noise will be increased in hose scenarios. The noise is correced by he paricle filer algorihm, which furher reduces he influence on he posiioning accuracy and improves he robusness of he algorihm in his paper. 5. Conclusion Due o he exising indoor posiioning echnology which has many deficiencies, such as high cos, low posiioning accuracy and low sensor uilizaion, an indoor pedesrian localizaion algorihm based on muli-sensor informaion fusion is proposed in his paper. The sensor buil-in smarphone can obain pedesrian s movemen informaion, such as acceleraion, sep size, direcion and so on. The proposed algorihm fused PDR and RSSI can improve he accuracy and he sabiliy. From he above experimenal resuls, he average error of he proposed mehod is above 0.35 meers in he range of 2m 8m, which is beer han he locaion algorihm based on Wi-Fi wih he average error of.65 meers. The proposed algorihm can improve he accuracy, sabiliy and real-ime of posiioning, and achieve good posiioning effec. In he proposed mehod, he way o hold he phone mus be consisen in order o ensure he accuracy of daa processing and collecion in he process of collecing fingerprin daa. Some furher work will be done o overcome he impac of he aiude angle of he phone on he indoor localizaion algorihm. Table 4. Robusness verificaion Normal Walk slowly Running he smar phone in pocke Muli-senor fusion 0.35 0.39 0.4 0.54 Wi-Fi.65.79 2.5 2.35 3
Acknowledgemen This work is suppored by he Naional Naural Science Foundaion of China (No.63707), he Guangxi Experimen Cener of Informaion Science (No.LD606X), he Guangxi Naural Science Foundaion (No.206GXNSFBA3804), and he China Posdocoral Science Foundaion (No.206M60292XB) and he Opening Projec of Guangxi Key Laboraory of UAV Remoe Sensing (No. WRJ206KF0). References [] Harle, R. (203) A Survey of Indoor Inerial Posiioning Sysems for Pedesrians. Communicaions Surveys & Tuorials IEEE, 5, 28-293. hps://doi.org/0.09/surv.202.292.00075 [2] Zhang, Z., Lu, Z., Saakian, V., e al. (204) Iem-Level Indoor Localizaion Wih Passive UHF RFID Based on Tag Ineracion Analysis. IEEE Transacions on Indusrial Elecronics, 6, 222-235. hps://doi.org/0.09/tie.203.2264785 [3] Zampella, F., Jimenez, R.A.R. and Seco, F. (203) Robus Indoor Posiioning Fusing PDR and RF Technologies: The RFID and UWB Case. [4] Cheng, L., Wu, C.D. and Zhang, Y.Z. (20) Indoor Robo Localizaion Based on Wireless Sensor Neworks. IEEE Transacions on Consumer Elecronics, 57, 099-04. hps://doi.org/0.09/tce.20.60886 [5] Yang, L., Chen, H., Cui, Q., e al. (205) Probabilisic-KNN: A Novel Algorihm for Passive Indoor-Localizaion Scenario. Vehicular Technology Conference. IEEE, -5. [6] Mao, G., Anderson, B.D.O. and Fidan, B. (2007) Pah Loss Exponen Esimaion for Wireless Sensor Nework Localizaion. Compuer Neworks, 5, 2467-2483. [7] Baala, O., Zheng, Y. and Caminada, A. (2009) The Impac of AP Placemen in WLAN-Based Indoor Posiioning Sysem. Eighh Inernaional Conference on Neworks. IEEE Compuer Sociey, 2-7. [8] Liao, L., Chen, W., Zhang, C., Zhang, L., Xuan, D. and Jia, W. (20) Two Birds wih One Sone: Wireless Access Poin Deploymen for Boh Coverage and Localizaion. IEEE Trans. Veh. Technol., 60, 2239 2252. hps://doi.org/0.09/tvt.20.209405 [9] Wang, C.S. and Lin, S.L. (205) An Inegraed Opimizaion Model for Wireless Access Poin Deploymen Consrucion, and Enhancemen. IEEE/ACIS Inernaional Conference on Sofware Engineering, Arificial Inelligence, NETWORKING and Parallel/Disribued Compuing. IEEE, -6. [0] Shin, H., Chon, Y. and Cha, H. (202) Unsupervised Consrucion of an Indoor Floor Plan Using a Smarphone. IEEE Transacions on Sysems Man & Cyberneics Par C Applicaions & Reviews, 42, 889-898. hps://doi.org/0.09/tsmcc.20.269403 [] Arvikar, S.A. (975) The Predicion of Muscular Load Sharing and Join Forces in he Lower Exremiies during Walking. Journal of Biomechanics, 8, 89-02. [2] Kim, J.W., Han, J.J., Hwang, D.H., e al. (2004) A Sep, Srideand Heading Deerminaion for he Pedesrian Navigaion Sysem. J Glob Posiion Sys., 3, 273-279. hps://doi.org/0.508/jgps.3..273 [3] Carpener, J., Clifford, P. and Fearnhead, P. (999) Improved Paricle Filer for Nonlinear Problems. 46, 2-7. [4] Damodaran, P. and Vélez-Gallego, M.C. (202) A Simulaed Annealing Algorihm 4
o Minimize Makespan of Parallel Bach Processing Machines wih Unequal Job Ready Times. Exper Sysems wih Applicaions, 39, 45-458. [5] Alasi, H., Xu, K. and Dang, Z. (2009) Efficien Experimenal Pah Loss Exponen Measuremen for Uniformly Aenuaed Indoor Radio Channels. Souheascon, SOUTHEASTCON '09. IEEE, 255-260. Submi or recommend nex manuscrip o SCIRP and we will provide bes service for you: Acceping pre-submission inquiries hrough Email, Facebook, LinkedIn, Twier, ec. A wide selecion of journals (inclusive of 9 subjecs, more han 200 journals) Providing 24-hour high-qualiy service User-friendly online submission sysem Fair and swif peer-review sysem Efficien ypeseing and proofreading procedure Display of he resul of downloads and visis, as well as he number of cied aricles Maximum disseminaion of your research work Submi your manuscrip a: hp://papersubmission.scirp.org/ Or conac jcc@scirp.org 5