Improvements of Indoor Fingerprint Location Algorithm based on RSS

Transcription

1 Internatonal Journal of Scence Vol.4 No ISSN: Improvements of Indoor Fngerprnt Locaton Algorthm based on RSS Quyue Zhu a, Qang Yu b, Q Lu c and Kun Sh d School of Computer and Software Engneerng, Xhua Unversty, Chengdu , Chna a quyue1316@163.com, b @qq.com, c @163.com, d @qq.com Abstract Wth the development of the Internet, the technology of locaton fngerprnt s wdely used because of the advantage n ndoor postonng performance of complex envronment. K neghbor algorthm based on sgnal strength s a common poston fngerprnt matchng algorthm. Ths paper frstly ntroduces the KNN fngerprnt localzaton algorthm, and ponts out the defcency of the algorthm. Then on the bass of the algorthm, correcton K neghbor algorthm s put forward through mproved the calculaton of Eucldean dstance. The smulaton results show that the mproved correcton algorthm can acheve good locaton accuracy compared to tradtonal KNN algorthm. Keywords Receved Sgnal Strength, Indoor Postonng, Locaton Fngerprnt, KNN Algorthm. 1. Introducton In recent years, the global postonng system (GPS) [1] and the technology of cellular wreless locaton has been able to provde a more accurate locaton servce outdoors, and has been appled n many felds such as mltary, traffc, surveyng and mappng. Compared wth the outdoor envronment, ndoor envronment s relatvely complex, and are more susceptble to nterference from the buldng and the multpath phenomenon because of ndoor sgnal reflecton, dffracton, refracton and scatterng. The exstng postonng technology can't meet the demand of relatvely accurate postonng ndoor. Therefore, loong for a nd of wreless postonng system whch has hgh precson and good stablty become urgent needs n the feld of ndoor postonng, and the localzaton algorthm s the ey to solve ths problem. Compared wth the tradtonal TOA [3], TDOA [4] and AOA locaton algorthm, ndoor postonng algorthm based on the receved sgnal strength [] (RSS) don't need to change the hardware equpment, and can realze postonng only by usng the exstng wreless networ resources, so t reduces the cost, and s wdely studed. At present, the ndoor locaton based on RSS nclude locatng method based on sgnal transmsson loss model and postonng method based on the poston fngerprnt. Sgnal transmsson loss method usually depends on that the sgnal transmsson loss n free space s nversely proportonal to the square of the dstance, because both the nfluence of the multpath effect and the dependence of the sgnal propagaton model mae ts accuracy restrcted. The locaton fngerprnt postonng technque depends on dentfyng the sgnal characterstcs of the target locaton, usng the dfferences of each pont sgnal to realze the localzaton. So the second method s less affected on the postonng accuracy n the complex ndoor envronment. However, complex and changeful of ndoor envronment tself wll lead to the volatlty of sgnal strength value as well as the external factors, so we must consder the effectve flterng of sgnal value and quc screenng method of processng poston fngerprnt data from a database to mprove the ndoor postonng algorthm, so as to mprove the accuracy of postonng. In the tradtonal fngerprnt locaton algorthm, we would get the fnal postonng result through by fngerprnt matchng or mappng usng RSS value of the reference nodes. At present, the common fngerprnt localzaton algorthm bascally has: the nearest neghbor method (NN), K neghbor 18

2 Internatonal Journal of Scence Vol.4 No ISSN: method (KNN), K weghted neghbor method, nave bayes algorthm. Ths paper bases on KNN algorthm to mprove the research.. Modfed Algorthm.1 KNN Algorthm KNN algorthm mprove the reference node number of postonng comparng wth the nearest neghbor method. Nearest neghbor method choose the reference poston whch s the most smlar to a locatng node at the sgnal strength vector as the estmate poston of the locatng poston. However, t s not feasble n practce due to the decson n the locaton node s too large and densty requrement of fngerprnt acquston wll be very hgh. The KNN algorthm mae up for t by selectng K (K ) fngerprnt data as locate reference ponts. Through calculatng and sortng eucldean dstance, the K fngerprnt data wth mnmum characterze the general regon of the locatng poston. Suppose there are n wreless access ponts (AP) and m reference ponts (RP) n postonng area. In the offlne phase, as shown n Fg. 1, we collect sgnals n each RP from dfferent AP and preprocess them to establsh a fngerprnt database. In the postonng stage, we frstly get sgnal nformaton of postonng node, namely, {(x,y)(rss1,rss,...rssn)}, then use KNN algorthm to match RPs. Fg.1 The establshment of the fngerprnt database The Eucldean dstance between locaton node and the reference ponts could be defned as: d n j1 R 19 S (1) j j Where = 1,,... m, j = 1,..., n. R j s the average sgnal strength namely fngerprnt nformaton whch locaton node receves from APj. Sj s the fngerprnt nformaton n the th reference pont from APj. So we can obtan reference poston whch have the mnmum value of Eucldean dstance by calculatng and sortng d. x, y 1 x, y x, y () 1 s the coordnates of the th RP. So we can calculate the estmated poston of locatng pont by type (). A ey factor affected the accuracy of postonng n KNN algorthm s to select the reasonable value of K. Ths value also relates to the densty of fngerprnt acquston, we can get t through several experments. However, the largest shortage of KNN algorthm s lacng n dstngushng the weght of fngerprnt, because the contrbuton of each RP to the anchor pont s dfferent and the mpact on matchng fngerprnts s not the same. Aganst the above, ths paper ntroduces weghted processng the standard devaton of sgnal strength to mprove the locaton method and puts forward modfed KNN algorthm.. Modfed KNN Algorthm. Because of the complexty of the ndoor envronment, the receved sgnal strength s not stable even at the same poston of dfferent tme, and the nstablty of sgnal s nstantated n the fluctuaton of

3 Internatonal Journal of Scence Vol.4 No ISSN: sgnal value. Consderng ths factor, Modfed KNN algorthm ntroduces standard devaton of sgnal strength when calculatng Eucldean dstance, and then gve a weght to each of the K fngerprnts accordng to the mproved Eucldean dstance, fnally estmate the locaton of anchor pont usng K weghted locaton. The mproved Eucldean dstance can be defned as: d ' j1 n R S j j j j (3) 1 N p1 N Rpj R j (4) Where = 1,,...m, j = 1,..., n, p = 1,,...N. N s the total number of measurements at each poston. R pj s the pth sgnal measurement from AP j at locatng poston. s the standard devaton of RSS at anchor pont. Accordng to the calculaton results of d set rght weght to K ponts on the bass of the sze of the j, we can sort them and select K mnmum ponts, and then d value. The smaller Eucldean dstance ndcates the greater smlarty between the reference pont and postonng pont, so we should allocate a larger weght, otherwse a smaller weght. So the locaton of locatng poston xy, can be calculated usng modfed KNN algorthm as follows: xy, w 1 1 x, y w Where = 1,,...,. x, y represents poston coordnates of the th RP from K nearest reference pont. s the weght of the th RP. Its computaton formula s as follows: w 1 w = d Judgng from ths, the above method could mprove the ndoor postonng accuracy compared wth the tradtonal KNN method. 3. Smulaton Results And Algorthm Analyss For evaluatng the performance of modfed algorthm, ths secton use Matlab to smulate and analyss. Smulaton area dagram s shown n Fg.. the smulaton area has 0 m * 15 m, whch contans 100 reference poston (*1.5 m ), and sets sx wreless access pont AP(randomly placed). (5) (6) Fg. The smulaton area To verfy that the mproved Eucldean dstance s better than the tradtonal Eucldean dstance, let us contrast two data collectons at the two same postons, as s shown n Table 1. 0

4 Internatonal Journal of Scence Vol.4 No ISSN: Table 1. Two sets of Eucldean dstance comparaton Rss_Avg1 Rss_Avg Rss_Avg1 Rss_Avg AP AP AP AP AP AP Tradtonal Improved In ths table, rss_avg1 and rss_avg respectvely are the average sgnal strength of two postons, the last two lnes respectvely express Eucldean dstance of two methods. From Table 1, we can see the fluctuaton of frst data set s larger. It llustrates that the mproved Eucldean dstance s bgger along wth the greater volatlty. Accordngly, the weght wll be smaller. In locatng phase, we respectvely use KNN algorthm and modfed KNN algorthm to match the fngerprnt database. The postonng accuracy of both methods are related to the selecton of K value, as s shown n Fg.3. The error of two algorthm decreases wth the ncrease of K value, and the postonng error of modfed KNN algorthm s obvous smaller than the tradtonal KNN algorthm. So t s that modfed KNN algorthm s obvously better than KNN algorthm. 4. Concluson Fg.3 The postonng error statstcs Ths paper frstly ntroduces the KNN algorthm, and then on ts bass put forward the modfed KNN algorthm through ntroducng the standard devaton of sgnal strength. Next, smulaton analyss was carred on. The expermental results show that the modfed KNN algorthm further mprove the precson of postonng. But defcency s that the worload to establsh a fngerprnt database s too heavy n offlne stage. So how to balance the fngerprnt ntegrty and great wor or fndng a fngerprnt automatc measurement method, s research content on next stage. Acnowledgements Ths paper s supported by Schuan Appled Basc Research Fund (The Research of Remote Intellgent Interacton Educaton Platform ) References [1] Kaplan, E.D.; Hegarty, C. Understandng GPS: Prncples and Applcatons, nd ed.; Artech House: Norwood, MA, USA, 006. [] KAEMARUNGSI K,KRISHNAMURTHY P. Properrtes of Indoor Receved Sgnal Strength for WLAN Localton Fngerprntng[C]//Proceedngs of the Frst Annual Internatonal 1

5 Internatonal Journal of Scence Vol.4 No ISSN: Conference on Moble and Ubqutous Systems: Networng and Servces (MobQutous 04). Massachusetts: IEEE, 004: [3] Alav B,Pahlavan K.Modelng of the TOA-based dstance measurement error usng UWB ndoor rado measurements[j].communcatons Letters,IEEE,006,10(4): [4] Curran K,Furey E,Lunney T,et al.an evaluaton of ndoor locaton determnaton technolages[j]. Journal of Locaton Based Servces,011,5():61-78 [5] Z.H. Ca, X. Xa, B. Hu, D.M. Fan: Improvement of ndoor sgnal strength fngerprnt locaton algorthm, Computer Scence, Vol.41 (014) No.11, p (In Chnese) [6] M. Yang, C.D. Xu, K. Zou, D.K. Yang.A Sensng Probablty-based Matchng Algorthm for WF Indoor Postonng Systems[J].Global Postonng System,013(6):7-11. [7] W.X.Zhang. The Fngerprnt Locaton Algorthm Based on RSSI of WF[D].Unversty of Electronc Scence and Technology of Chna.015.