Experiments for Leveled RFID Localization for Indoor Stationary Objects

Transcription

1 th Interntionl Conferene on Informtion Tehnology: New Genertions Experiments for Levele RFID Loliztion for Inoor Sttionry Ojets Mtthew Chn Computer Systems Tehnology Dept. NYC College of Tehnology, CUNY 300 Jy Street, Brooklyn, NY Xiowen Zhng Computer Siene Dept. College of Stten Isln, CUNY 2800 Vitory Blv, Stten Isln, NY Astrt Bse on the multiple power levele pssive RFID system Chn n Zhng [21] propose, there re three innovtive inoor loliztion lgorithms, nmely the levele nerest-neighor, levele multiltertion n levele Byesin inferene, to lote sttionry ojets. In orer to exmine these lgorithms in relisti irumstne, we rry out experiments in lortory environment. The hrwre of the RFID system in use is the Alien Tehnology evelopment kit. The experimentl results re onsistent with the outomes from the omputer simultions. We re le to onfirm tht the verge loliztion error results for these lgorithms re quite smll. Keywors-RFID loliztion; experiment; nerest-neighor; multiltertion; Byesin inferene I. INTRODUCTION As the ost n size of the Rio Frequeny Ientifition (RFID) tgs n reers hve een eresing rmtilly over the pst two ees, their impt on the utiliztion of the tehnology in the emi [1], inustril n ommeril setors hs een inresing tremenously. The pility n pity to ientify, store n retrieve lrge mount of t emee in these tiny tgs, ouple with the spee n ury to etermine the ojet s estimte lotion, mke RFID tgs muh more powerful n effiient, ompre to the tritionl tehnology tht requires physil ontt n line-of-sight suh s roes. Most of the urrent si RFID systems re ple of proviing orse-grine lotion informtion out tgge ojet; they n tell whether the ojet is present or sent in the proximity. However, more urte lotion informtion of ojets oul enhne these pplitions, n enle lrge numer of other pplitions. For exmple, in lrge wrehouse, lrge lirry, hospitl, or smrt home, how to quikly n urtely lote ertin ojets, ooks, or ptients, in ost-effetive mnner, is highly esirle. Loting n positioning of ojets n persons n e highly influentil n enefiil. Lolizing tehniques n e lssifie into RF-se (Rio Frequeny) n non-rfse [2]. GPS (Glol Positioning System) is suessful RF-se metho for positioning outoor ojets. However, GPS reeivers in generl o not work in inoor settings euse the stellite signls re hevily ttenute y the uiling wlls n floors. RF-se loliztion pprohes lso inlue wireless lol re network (WLAN), wireless sensor network (WSN), n RFID loliztions. In WLAN loliztion, WLAN evie n e lolize se on eon signl strength informtion etween the evie n ess points. Some of the mjor stuies on WLAN loliztion n e foun in [3-5]. Generlly speking, WSN loliztion tehniques re se on AOA (ngle-of-rrivl), TDOA (time-ifferene-of-rrivl), n RSS (reeive signl strength). WSN loliztion is n importnt re of reserh n mny stuies hve een reviewe in [6-8]. RFID loliztion tehniques re surveye in [2, 9-10]. They re similr to other RF-se loliztions in priniple; the min vntge over WLAN n WSN pprohes is the inexpensiveness n ese of eployment [2]. We will ress RFID loliztion in etil in the susequent setions. Non- RF-se loliztion tehniques employ ousti, vision, ultrsoun, infrre n lser s mesuring signl. These tehniques typilly re uilt on time-of-rrivl (TOA) n tringultion or multiltertion lgorithms. A. RFID loliztion systems There hve een severl loliztion shemes tht re highly influentil n well ite in the literture. A preeminent pioneer in the loliztion of sttionry ojets in n inoor environment employing the RFID tehnology with the eployment of referene tgs is the Lotion Ientifition se on Dynmi Ative RFID Clirtion (LANDARC) system of the LANDMARC sheme [11]. To improve the ost-effetiveness of the system, the metho tkes vntge of the reltively eonomil RFID tgs, inste of using the extr more ostly reers. Utilizing the RSS of the etete fixe lotion referene tgs, ouple with the k-nerest neighor (knn) lgorithm, the estimte lotion of the sttionry ojet n e ompute. Moreover, the sheme hs een eploye n investigte in serious infetious isese setting uring the SARS (severe ute respirtory synrome) outrek to filitte the rpi time requirement of loting ptients n the trking of the isposl of meil wstes [12]. The SpotON [13] ustom uilt tgs enhne the onveniene, implementtion, n ury of n RFID loliztion system. The min ie is tht severl reers from ifferent fixe positions will reeive the RSS mesurements from the etete SpotON tg, thus their estimte istnes from it. The metho lims tht the loliztion error n e well within the onemeter rnge /14 $ IEEE DOI /ITNG

2 In new or unknown inoor strutures or uilings, trustworthy guie og most likely will not perform well ompre to the propose rooti guie RG [14] whih is lin user to trverse the ifferent n strnge enlose environments. Pssive RFID tgs re ple throughout the environment. They will lert n inform the system of the urrent lotion of the user n lso vise him where n whih iretion he shll ontinue to trvel. The RG system tkes vntge of the empty spes in the potentil fiel where the lin user n sfely trverse. One of the min pplition res of the wireless lol positioning system (WLPS) [15] is the voine of ollision for trveling utomoiles. Eh DBS (Dynmi Bse Sttion) in the system ontinuously etets, monitors, trks n lotes moving vehiles in its overge re, whih n e signifintly enlrge y employing the tehnology of multi-hop loliztion. By tthing tive RFID tgs onto the vehiles, their estimte lotion n e ompute using the iretion-of-rrivl n time-of-rrivl of the reeive rio signls. Both the first n the seon versions of R- LIM (RFID-se lirry informtion mngement) ttempt to tkle the hllenge of pinpointing the lotion of ny ooks in rel time in lirry environment, in n eonomil n fesile mnner [16-17]. The ging-ounter pproh, se on the proximity n nerest neighor resoning, is pplie to R-LIM to omplish the tsk of loting lirry ooks. Generlly speking, the implementtion of RFID loliztion systems n e lssifie into two kins of pprohes: The first lss of shemes suh s RSS s well s TOA employ the propgtion moels of RF to lulte the estimte istne. On the other hn, the seon lss of pprohes oes not tke vntge of the RF propgtion moels. Aitionlly, the seon lss n e ivie further into two tegories: (1) RF istriution lirtion followe y the estimtion of lotion, for exmple, Byesin inferenes n multiltertion; (2) estimtion of lotion iretly, for exmple, kernel-se lerning, proximity, n nerest neighors. We orgnize the rest of the pper s follows. Setion 2 provies rief introution to the three pssive RFID levele inoor loliztion lgorithms n the outomes of omputer simultions. Setion 3 presents the isussion of the settings of experiments for the levele RFID loliztion lgorithms. Setion 4 inlues the results n nlysis of the experiments. Setion 5 ompres our methos with other RFID loliztion lgorithms. Finlly, the onlusion of the pper is rrie out in the lst setion. II. LEVELED PASSIVE RFID LOCALIZATION METHODS Severl pssive RFID loliztion shemes for inoor fixe entities were put forwr y Chn n Zhng [18]. The lgorithms re ientifie s levele nerest-neighor (LNN), levele multiltertion (LML) n levele Byesin inferene (LBI). These methos employ the etetle or non-etetle ounts of the entities s the RSSI (reeive signl strength inition). The propose pprohes n lulte with ury the trget's lotion se on the results of the omputer simultion. A. The Three Algorithms The levele nerest-neighor (LNN) lgorithm utilizes the omintion of the multi-levele etetle ounts n the nerest-neighor metho [11] to reh etter lotion ury. The RFID reer with three istint power levels high, meium n low res pssive referene tgs tht re ple onto the floor in gri formt (Figure 1). The referene tgs, inite y lk ots, re ple s known oorintes, with the soli-line squre s the trget. Figure 1 emonstrtes how the LNN metho works with the three power levels: the lrge irle on the left of the figure illustrtes the use of the RFID reer t high power level, the smll irle on the right t low power level, n the mile irle illustrtes meium power level. The referene tgs,, n, within the lrgest irle, re the four nerest neighors when the high-level power is engge. The enter of, represente y the lk sh-line squre, is the urrent estimte lotion of the trget. When we just the power level to meium, the reer is le to etet two nerest neighors, n, s shown within the re sh-line irle. The re sh-line squre, the new estimte lotion of the trget, is otine, ssuming simplisti lultion, y moving the first estimte from the high-level power to the right of the hlf istne of the enter of to the line. Lstly, we lter the power level to low; the reer is le to etet only one nerest neighor, the referene tg. Aoringly, the sheme shifts the seon estimte, otine fter the meium power level, own hlf istne of the enter of towr the line. Hene, we re le to otin the finl estimte lotion of the trget, the lksoli line squre. Consequently, using the LNN tht omines the nerest-neighor metho n the multi-levele etetle ounts, the estimte lotion of the trget eomes inresingly more preise. The LNN sheme is lso weighte verge metho [19] in the sense tht s the referene tgs re loser to the trget, the weight they get will e inrese. The following formul expresses the estimte s In the exmple epite in Figure 1, we hve N = {,,, } s the set of neighors. Hene, the oorintes of the referene neighors re (x i, y i ) where i =,,, or. The weight w i is equl to the quntity of times tht referene tg hs een etete. Given tht the tgs n hve only een etete one, hene w = w = 1; tg etete twie, therefore w = 2; n tg etete thrie, so w = 3. Figure 1. Levele nerest-neighor lgorithm y three levels. 164

3 A well estlishe sheme to quire the whereouts of ojets is the multiltertion pproh. On twoimensionl plne, given t lest three referene points with known oorintes n their respetive istnes to the trget, then trget s lotion n e sertine. For exmple, let there e n referene points (or eons) with known istnes: = ( x x1) + ( y y1) = ( x x2) + ( y y2) n = ( x xn) + ( y yn) where i is the istne etween the trget n eon i, (x i, y i ) is the oorintes of the eon, n (x, y) is the trget's position. To linerize the system of the n equtions ove, one will sutrt ny single one eqution from the rest of the other equtions. It is ommon to just use the first eqution to sutrt from the others. Therefore, we hve: ( x1 x2) x+ ( y1 y2) y = 21 ( x1 x3) x+ ( y1 y3) y = ( x1 xn) x+ ( y1 yn) y = n1 In mtrix form, the equtions eome: M T =, where x1 x2 y1 y2 21 x1 x3 y1 y3 x 31 M =, T =, n = y... x 1 xn y1 yn n1 Usully n (the numer of referene points) is lrger thn three, so the system of equtions eomes over-etermine. To otin the oorintes of the trget noe, one trnsfers the equtions to liner lest squres prolem: Min M T 2. The metho of multiltertion hs to first lirte the istriution of RF signl. For LML, the proess of lirtion involves quiring the weights, i.e., the etetle ounts. A more potent signl strength will e inite y lrger numer of etetle ounts. We use the inverses of orresponing squre weights to reple the istnes from ll the referene tgs in N tht re etete. Therefore, we n rewrite the system of equtions for the metho of generl multiltertion [2] s in eqution (2), where the oorintes of referene tg re represente y ( x, y ). For the metho of LBI, the posterior proility of the trget entity t lotion t nees to e estimte reursively. In the se s shown in Figure 2, the strengths of the mesures s result of the etetle of referene tgs (,,, ) re s 1 = (1, 1, 1, 1), s 2 = (0, 1, 1, 0), n s 3 = (0, 0, 1, 0), where 0 inites sene n 1 signifies presene of the orresponing referene tg. As the system is presume to e proess of Mrkov, thus for the t tht is given, the proilities of s i re inepenent of one nother. From the series of mesurements provie y the strengths of its neighors, the trget entity's proility t lotion t n e ompute using the following reursive eqution where the ftor of normliztion is n the proility of signl strength of mesurement i given the trget t is speifie y p(s i t). For solving the reursive Eqution 3, we nee to etermine the priori proility p(t s 0 ) p(t) for the trget t, whih represents ll ville informtion known out the trget eforehn. As we only know the ounries out the trget in vne, we will use the uniform istriution proility for p(t s 0 ) sine it is the lest ise [2]. Therefore, p(t s 0 ) = p(t) = 1/N, where N is the totl numer of referene tgs in the region of the trget tg. In our levele lgorithm, we mke three mesures, i.e., n = 3. If we tke ount of the initil s 0, the Eqution 3 will eome p(t s 0, s 1,..., s 3 ) = p(s 3 t)p(s 2 t)p(s 1 t)p(t). (4) In the se of Figure 2, N = 9 n we get the following equtions for eh referene tg: α p ( t s, s,..., s ) = α p( s t) p( s t) p( s t) p( t) = x x x ( 5) ( 0, 1,..., 3) = α ( 3 ) ( 2 ) ( 1 ) ( ) = x x x N ( 6) ( α 0, 1,..., 3) = α p( s3 t) p( s2 t) p( s1 t) p( t ) = x x x N ( 7) (,,..., ) = α ( ) ( ) ( ) α ( t ) = x x x ( 8) p t s s s p s t p s t p s t p t p t s s s p t s s s p s t p s t p s t p Then, we otin = 126 from the following normliztion eqution p + p + p + p = ( )/(7 x 6 x 4 x 9) = 12 /1512 = 1. (9) Therefore, we hve p = p = 126 x 1/1512 = 0.083, p = 126 x 4/1512 = 0.33, p = 126 x 6/1512 = t s 1 s 2 s 3 N α Figure 2. Levele Byesin inferene y three levels. N 165

4 B. Silmultion Results In orer to onut simultions of the three methos, LNN, LML n LBI, omputer progrm ws evelope [18]. We ompute the lotion errors of these levele loliztion methos se on trget with vrious referene tg ensities n rius rtios of the etetion rnges of the high-level power s perentge of the efult R (rius). For the purpose of the simultion stuy, the efult R is efine s the rius of the high-level power etetion rnge when the ensity of referene tgs is t 4 meters. The rtio of the rii of the set of three etetion rnges of the high, meium n low power is 4:3:2, whih mens the high power etetion rnge is twie of the low power one while the meium one is the verge of the high n the low ones. The ensities of referene tgs rnge from one to five meters with one-meter intervl. On the other hn, the rius rtios of the etetion rnge (of the high-level power) vry from 80% to 150% of the efult R. Figure 3 illustrtes the outomes of the simultion with three meter referene tg ensity. Plese refer to [18] for further etils. Lotion error (in meter) Figure 3. III. 80% R 90% R 100% R 110% R 120% R 130% R 140% R 150% R High-level power etetion rnge s perentge of the "efult R" LNN, LML n LBI simultion results t 3-meter tg ensity for ifferent rius rtios. EXPERIMENTAL SETTINGS FOR THE LEVELED RFID LOCALIZATION METHODS In orer to exmine the propose methos in relisti irumstne, we eie to rry out experiments in the environment of lortory. The reer we hose to employ is the ultr-high frequeny RFID reer me y Alien Tehnology (ALR-9900+), together with pir of the Alien ALR-8696-C ntenns; n the pssive RFID tgs use re Alien type G tg (ALN-9654). A. Sensitivity n Orienttion of RFID Tgs As other reserhers hve inite, tg sensitivity n qulity is n importnt issue when one ttempts to implement RFID loliztion sheme. We exmine every tg tht we hve quire to wee out the outliners, those tht re very sensitive or too insensitive. Bsilly we hoose the tgs tht elong to the mile of the pk tegory; i.e. they hve norml n verge sensitivity for tht type of tgs we re using. In generl, the est orienttion of tg for the purpose of etetion is tht it fes the ntenn in prllel fshion. But LNN LML LBI s the ntenn is normlly stning vertilly, it is highly not visle to lso hve the tg stning up s tht kin of setup will likely interfere with the movement of n entity when moile loliztion is require to e implemente. Hene the next est pproh is to hve the tg lie flt on the floor ut in prllel orienttion in referene to the footprint of the ntenn. But sine the tgs re ple ll over the floor of the testing re in squre gri pttern, n even with four ntenns use, there is no gurntee tht tg whih is within the etetion rnge of n ntenn will e prllel to n ntenn. To enhne the likelihoo tht the etetle referene tg s orienttion is t zero egree from n ntenn, we put two tgs in ross pttern t every single referene point in orer to inrese the etetility of tht referene point. B. Experimentl Setup The proility of etetion of tg in the qurter-ish region iretly in front of the ntenn is lmost very high, usully etween 85% to 100%, i.e. if n RFID tg is in tht region not too fr from the ntenn n when the reer sns for the tg, the suessful re ounts out of the 100 sns is t lest 85. To ensure tht the full irulr region (i.e. 360 egrees) roun the trget hs similr etetion likelihoo n pility, we nee to employ four ntenns t 90 egrees to eh other, single ntenn turning 90 egrees four times, or two k-to-k ntenns turning 90 egrees one for snning. We elet the lst option s it is the lest umersome to implement n exeute. To ensure tht the surfe is highly len, free of ust or eris, we eie to put rn new pper poster ors on the floor efore we ly own the pssive RFID tgs with the pir of RFID ntenns ple t the enter of the testing re within the lortory room. The tgs re ple on the floor in squre gri pttern with eh ross pttern tg pir positione hlf meter from one nother (see Figures 4-6). The size of the testing re in the lortory tht is use for the experiments is forty two n qurter squre meters ( squre region of sq. m.). We o not ple the RFID tgs on the testing region s ounries; hene the first pir of ross tgs re li t the lotion of x, y oorintes of 0.5, 0.5 (meter) while the lst pir is t 6.0, 6.0 (meters). All together, one hunre n forty four pirs of RFID tgs re ple in the testing region with the trget (i.e. the pir of k-to-k ntenns) lote t the enter of the re (x, y = 3.25, 3.25). Figure 4. Shemti top view of the enter of the testing re. 166

5 Sine the true tul lotion of the trget in the experiments is t 3.25, 3.25 meters, we hve the following results (errors in istne of estimte trget lotion) n lso the verge of the four ses for the three multi-levele methos (see Tle II): TABLE II. ESTIMATED LOCATIONS ERRORS (IN METER) Figure 5. Setting up the testing re with the ntenns t the enter. Figure 6. The testing region with 144 pirs of RFID tgs whih re hlf meter prt. Through the justment of the power ttenution of the RFID reer, we n ontrol the etetion rnge of the reer for the tgs tht re eing querie. Our pprohes stte tht there re three etetion rnges: lrge, meium n smll; n the rtio of those rnges use in our simultions is 4:3:2. In the experiments, we test four instnes with the sme rtio for the three etetion rnges: i. 1.0 : 0.75 : 0.5 meter, ii. 2.0 : 1.50 : 1.0 meter, iii. 3.0 : 2.25 : 1.5 meters, iv. 4.0 : 3.00 : 2.0 meters. IV. EXPERIMENTAL RESULTS AND ANALYSIS Bse on the experiments n the results of omputtion, the estimte lotions (x, y oorintes in meters) of the trget for the three levele methos (Levele Nerest- Neighor, Levele Multiltertion n Levele Byesin Inferene) re shown in Tle I: TABLE I. ESTIMATED LOCATIONS OF THE TARGET Cse # LNN LML LBI i , , , ii , , , iii , , , iv , , , Cse # LNN LML LBI i ii iii iv verge (of 4 ses) Generlly speking, it is pprent tht the results of the propose methos of loliztion experiments in the lortory setting re onsistent with tht of the forementione results one in the omputer simultions. Moreover, the verge results for oth the LNN n LBI ( n meter) re quite goo. Even the LML s verge result of meter is rther resonle, given tht the size of n inoor spe n e extremely lrge (for exmple squre meters). Finlly, for the four ses n the three methos use, the est n the seon est results of n meter (Levele Byesin Inferene se i n Levele Nerest-Neighor se iii) re truly impressive. V. COMPARISON OF LEVELED METHODS WITH OTHER RFID LOCALIZATION ALGORITHMS In this setion, we present the omprison of our multiple power levele methos with other pprohes from the literture (see Tle III). Both LANDMARC [11] n our shemes employ fixe lotion referene tgs tht re eploye on the floor spe of n inoor l environment. The optiml result of verge estimte istne errors tht LANDMARC otine is 1.09 meters, with referene tg ensity of single tg per squre meter, n fixe numer of four referene tgs in the omputtion. The verge estimte istne errors of our three methos re etween n meter (see Tle II), with referene tg ensity of one hlf meter, i.e. there re up to nine pirs of tgs in squre meter of floor spe. The numer of referene tgs use in the omputtion of our methos my vry in every single run, s we use three ifferent power levels of the RFID reer, whih generte three ifferent etetion rnges (long, meium n short). Bsilly, for the four ses teste in our experiment, the etetion rnges vry from reltively short istne of one hlf meter to meium-long istne of four meters (see Setion III). One possile explntion for the omprtively less stisftory performne of the LANDMARC sheme versus ours is tht, even though the LANDMARC pproh relies 167

6 on RSS mesurements, RSS n only e otine iniretly through the justment of the eight power levels of the reers, s expliit RSS reings were not provie y the reers t tht time. Furthermore, urrently ville RFID reers provie mny more isrete power levels (150 for ours versus 8 of LANDMARC's). SpotON [13] ttempts to otin oth the reltive positionl informtion n solute lotion t of tgge entities. The former ojetive n e hieve if ll the relevnt entities re tgge n the estimte istnes mong them n e mesure n ompute urtely. In orer to otin the solute positionl informtion of the trgets, tgs with fixe known positions must e eploye. Our methos entil the use of trgets with the RFID reer ntenns tthe to them, while SpotON hs tive RFID tgs ffixe to the trget entities. SpotON n our LML re similr in the sense tht oth use ltertion in their lultion, i.e. from the estimte istnes mong the tgs or etween the tgs n the reer, one n ompute the pproximte true lotion of the trget entity. SpotON's lime ury of one meter is resonly goo, though our ompute ury of 0.34 meter using the LML metho is signifintly etter. One most likely ftor ontriuting to our inrese ury is the high ensity of referene tgs employe. "knn-plus" [20] is n improve RFID loliztion system se on the lssi knn pioneere y LANDMARC. Two novel itionl tehniques introue to enhne the ury of lotion estimtion of sttionry trget re RSSI orretion n superfluous referene tgs purging. knn-plus oserves tht the isrepny of RSSI vlues mong the tgs n e quite lrge even uner norml uniform onitions. To ensure ury, very lrge numer of smples of RSSI vlues (twenty thousn) for eh tg re mesure uner ientil onitions n the verge RSSI vlue for eh tg is use for the orretion proeure. Therefore, the RSSI-orrete figure for speifi referene tg is the verge RSSI vlue of tht tg ivie y the RSSI isrepny rtio of the sme referene tg over the trget tg. For exmple, if the verge RSSI of referene tg is e-06 mw n its isrepny rtio is 1.89, then its RSSI-orrete vlue is e-06 mw. In ition, knn-plus lims tht ny tgs tht re physilly lote istnt from eh other ten to hve very ifferent reilities. Therefore, etete referene tgs tht re fr from the trget tg's nery neighorhoo my not provie useful guine for etermining the ext lotion of the trget n my inste introue noise n loliztion estimtion error. As result, their exlusion my e enefiil to the ury of the loliztion sheme. For exmple, if the trget tg is situte t istne of two n hlf meters from the RFID reer ntenn n the rtio of snning rte etween the trget ojet's tg n the surrouning referene tgs is within fifty one perent whih lso hppens to e the stnr evition numer from their experiments then those referene tgs shoul not e inlue in the omputtion of the trget position when using knn, s their extreme issimilrity in reility will rener them ineffetive in etermining of the urte lotion of the trget ojet. By opting the originl knn pproh with their performne enhnement n error reution tehniques, knn-plus is le to otin loliztion error of only meter in their experiments. It is pprent tht the meter performne of our LML metho is omprle to tht otine y knn-plus. Virtul Referene Elimintion (VIRE) [21] is lso se on the sheme use y LANDMARC, with the introution of two innovtive tehniques: the use of virtul tgs to signifintly inrese the ensity of the tul tive RFID referene tgs n the exlusion of improle trget lotions. For the usge of virtul tgs, eh rel gri ell overe y four tul tive RFID referene tgs is prtitione into n n ientilly size virtul gri ells. One n presume every virtul gri ell is overe y four virtul RFID referene tgs. One enefit of using virtul tgs is ost svings s utilizing lrge quntities of tive RFID tgs n e expensive. Moreover, itionl tul tgs my proue exessive rio wve interferene whih my negtively impt the ury of RSSI reeive. The RSSI figures of the virtul tgs n e esily interpolte linerly from the RSSI vlues of the tul tgs. Regring the exlusion of improle trget lotions, sine multiple reers re eploye, eh one will provie the RSSI numers of the trget tg n its neighor tgs when the trget is in the etetion rnge. Therefore, the etete neighorhoo regions of eh reer my vry. By merging these regions together, n exluing regions tht re elow the speifi threshol, one n otin more preise reing of the true lotion of the trget, s the unlikely position regions re eliminte. Bsilly, RSSI eviene from multiple reers is to e onsiere wheres onfirmtion from only single reer shll e exlue. Experiments on three environmentl settings re rrie out n the men loliztion error of them is 0.20 meter. The tive sheme of the Simplex metho [22] uses n RFID reer tht is tthe to or mounte on trget. Referene tgs re eploye oth on the floor n the eiling of the room. Sine the snning rnge of the reer is fixe istne, the etete referene tgs shoul form two irles (on the floor n eiling). The estimte trget lotion n e otine y omputing the rii n enters of the irles. Moreover, eh irle n e forme from two orers: inner noes versus outer noes. The inner noes re the etete referene tgs tht re on the orer while the outer noes re the non-etete tgs tht re jent to the etete ones. Bse on simultions, it n e onlue tht using oth orers in the omputtion genertes etter results. When using sets of referene tgs on the orers, in orer to ompute the enter of the irle n the rius, one hs to minimize the error ojetive funtion y pplying the Neler-Me Simplex metho. Experiments re performe for the following three senrios: trget t the enter of room with single ntenn tthe to the reer; rottions of the ntenn to emulte the positive influene of n ntenn rry; n trget move to 168

7 ner the wlls n lose to the orner of the room. The outomes otine re 1.691, 0.260, n feet respetively. If we only onsier the outomes of the ntenn rry ses, the men lotion error is meter (= ( )/2 = foot). Compre to our LML metho, the ury of the Simplex sheme is similr. TABLE III. COMPARISON OF RFID LOCALIZATION ALGORITHMS Loliztion Algorithm LAND- Positioning Sheme Aury (in meter) Refer -ene Tgs Tg Type MARC knn 1.09 yes tive SpotON RSS ltertion 1.0 no tive knn-plus knn yes pssive VIRE knn 0.20 yes tive Simplex knn optimiztion yes tive LML ltertion yes pssive LNN LBI nerest neighor yes pssive Byesin inferene yes pssive VI. CONCLUSION As onfirme y the experimentl results in lortory environment, the levele etetle ount RFID loliztion methos generte urte performne. Sine not ll tgs of the sme type re rete equl in terms of sensitivity, we hve to pre-selet qulifie tgs for our experiments. The setup of the experiment n the pre-seletion of qulifie tgs re the most time-onsuming. We exmine four tul ses of loliztion of sttionry ojets in the experiments. The outomes of the four ses re lso nlyze. Our stuy inites tht loliztion error of less thn three n hlf entimeters is ttinle. ACKNOWLEDGMENT This work ws supporte in prt y the PSC-CUNY Reserh Awr REFERENCES [1] M. Kritzler, J. Stephnie, K. Philipp, n K. Antonio. Inoor trking of lortory mie vi n RFID-trking frmework. In Pro. of the first ACM int l workshop on Moile Entity Loliztion n Trking (MELT 2008) in GPS-less environments. [2] J. Zhou n J. Shi. RFID loliztion lgorithms n pplitions review. J. of Intelligent Mnufturing, 20: pp , [3] P. Cstro, P. Chiu, T. Kremenek, n R. Muntz. A proilisti room lotion servie for wireless networke environments. In Pro. of Uiomp 01, pp , [4] A. Smligi n D. Kogn. Lotion sensing n privy in ontext-wre omputing environment. IEEE Wireless Communitions, 9(5):10-17, [5] K. Yu, J. Montillet, A. Rhin, P. Cheong, n I. Oppermnn. UWB lotion n trking for wireless emee networks. Signl Proessing, 86: , [6] 7. A. Boukerhe, H. Oliveir, E. Nkmur, n A. Loureiro. Lotion systems for wireless sensor networks. IEEE Wireless Communitions, 14(6):6-12, [7] 8. G. Mo, B. Fin, n B. Anerson. Wireless sensor network loliztion tehniques. Computer Networks, 51: , [8] 9. N. Ptwri, J. Ash, S. Kyperounts, I. Hero, A. Moses, n N. Correl. Loting the noes: oopertive loliztion in wireless sensor networks. IEEE Signl Proessing, 22(4):54-69, [9] 10. M. Bouet n A. Sntos. RFID tgs: positioning priniples n loliztion tehniques. In Pro. of IFIP Wireless Dys, pp. 1-5, Dui. [10] 11. J. Hightower n G. Borriello. Lotion systems for uiquitous omputing. IEEE Computer, 34:57-66, [11] L. Ni, Y. Liu, I. Lu, n A. Ptil. LANDMARC: Inoor lotion sensing using tive RFID. Wireless Networks, 10(6): , [12] C. Li, L. Liu, S. Chen, C. Wu, C. Hung, n X. Chen. Moile helthre servies system using RFID. In Pro. of 2004 IEEE Int l Conf. on Networking, Sensing & Control, volume 2, pp , [13] J. Hightower, C. Vkili, G. Borriello, n R. Wnt. Design n lirtion of the SpotON -ho lotion sensing system. Teh. Rep., Computer Siene Dept., Univ. of Wshington, Aug [14] V. Kulyukin, C. Ghrpure, J. Niholson, n S. Pvithrn. RFID in root-ssist inoor nvigtion for the visully impire. In Pro. of 2004 IEEE/RSJ Int l Conf. on Intelligent Roots n Systems, pp , [15] H. Tong n S. Zekvt. A novel wireless lol positioning system vi merger of DS-CDMA n emforming: proility-ofetetion performne nlysis uner rry perturtions. IEEE Trns. on Vehiulr Tehnology, 56(3): , [16] J. Choi, D. Oh, n I. Song. R-LIM: n fforle lirry serh system se on RFID. In Pro. of the 2006 Int l Conf. on Hyri Informtion Tehnology (ICHIT), pp , [17] J. Choi n D. Oh. Tg-only ging-ounter loliztion for the R- LIM2 system. J. of Informtion Systems Frontiers, [18] M. Chn n X. Zhng. Levele inoor loliztion lgorithms se on pssive RFID. In Pro. of the 2012 Int l Conf. on Wireless Networks (ICWN 12), Ls Vegs, July 16-19, 2012, pp [19] A. Pppostolou n H. Chouhi. RFID-ssiste inoor loliztion n the impt of interferene on its performne. J. of Network n Computer Applitions, 34: , [20] J.S. Choi, H. Lee, R. Elmsri, n D.W. Engels. Loliztion systems using pssive UHF RFID. Fifth Interntionl Joint Conferene on INC, IMS n IDC, [21] Y. Zho, Y. Liu, n L.M. Ni. VIRE: Ative RFID-se loliztion using virtul referene elimintion. In Pro. of ICPP, [22] C. Wng, H. Wu, n N.-F. Tzeng. RFID-se 3-D positioning shemes. In Pro. of INFOCOM, pp ,