Journl of Globl Positioning Systems (2004) Vol. 3,. 1-2: 273-279 A Step, Stride nd Heding Determintion for the Pedestrin Nvigtion System Jeong Won Kim GNSS Technology Reserch Center, Chungnm Ntionl University, Kore e-mil: kimjw@cnu.c.kr Tel: + 82-42-821-7709 ; Fx: +82-42-823-5436 Hn Jin Jng GNSS Technology Reserch Center, Chungnm Ntionl University, Kore e-mil: hndoo01@cnu.c.kr Tel: + 82-42-821-7709 ; Fx: +82-42-823-5436 Dong-Hwn Hwng GNSS Technology Reserch Center, Chungnm Ntionl University, Kore e-mil: dhhwng@cnu.c.kr Tel: + 82-42-821-5670 ; Fx: +82-42-823-5436 Chnsik Prk School of Electricl nd Computer Engineering, Chungbuk Ntionl University, Kore e-mil: chnsp@cbucc.chungbuk.c.kr Tel: + 82-43-261-3259 ; Fx: +82-43-268-2386 Received: 15 v 2004 / Accepted: 3 Feb 2005 Abstrct. Recently, severl simple nd cost-effective pedestrin nvigtion systems (PNS) hve been introduced. These systems utilized ccelerometers nd gyros in order to determine step, stride nd heding. The performnce of the PNS depends on not only the ccurcy of the sensors but lso the mesurement processing methods. In most PNS, verticl impct is mesured to detect step. A step is counted when the mesured verticl impct is lrger thn the given threshold. The numbers of steps re miscounted sometimes since the verticl impcts re not correctly mesured due to inclintion of the foot. Becuse the stride is not constnt nd chnges with speed, the step length prmeter must be determined continuously during the wlk in order to get the ccurte trvelled distnce. Also, to get the ccurte heding, it is required to overcome drwbcks of low grde gyro nd mgnetic compss. This pper proposes new step, stride nd heding determintion methods for the pedestrin nvigtion system: A new relible step determintion method bsed on pttern recognition is proposed from the nlysis of the verticl nd horizontl ccelertion of the foot during one step of the wlking. A simple nd robust stride determintion method is lso obtined by nlysing the reltionship between stride, step period nd ccelertion. Furthermore, new integrtion method of gyroscope nd mgnetic compss gives relible heding. The wlking test is preformed using the implemented system consists of 1-xis ccelerometer, 1-xis gyroscope, mgnetic compss nd 16-bit microprocessor. The results of wlking test confirmed the proposed method. Key words: Pedestrin nvigtion system, Step detection, Stride determintion, Heding determintion 1 Introduction Pedestrin nvigtion system(pns) provides velocity nd position of person nd cn be pplied to mny other res such s E-911 service, loction bsed services (LBS), tourism, rescue, militry infntry, medicl studies, leisure, nd nvigtion for the blind. In PNS, it is necessry to locte the position of the user in ny time nd ny environment. Even GPS is useful personl nvigtion system, its vilbility is significntly reduced when signl is blocked. Also ultr wide bnd (UWB) nd rdio frequency identifiction (RFID) techniques re introduced for personl nvigtion, but these systems
274 Journl of Globl Positioning Systems require dense infrstructure. For these resons, selfcontined nvigtion system bsed on ded reckoning (DR) principle is of interest (B. Merminod et l, 2002). To locte the position of the PNS user, distnce nd heding from known origin hve to be mesured with n cceptble level of ccurcy. In PNS, n ccelerometer is used to count the number of steps, which is combined with the stride to obtin the trvelled distnce. In ddition, mgnetic compss or gyroscope is used s heding sensor. The stride nd step re importnt prmeters for PNS ded reckoning lgorithm. Mny methods hve been suggested to detect step. One such method is to detect the peks of verticl ccelertion, which correspond to the step occurrences becuse the verticl ccelertion is generted by verticl impct when the foot hits the ground. If the verticl impct is lrger thn given threshold, it is considered s step. Since the pttern of impct signl depends on type of movement (going up or down stirs, crwling, running etc.) nd type of ground over which the person wlks (hrd or soft surfce, snd), the determintion of threshold is not so esy for relible step detection (Ldetto nd Merminod, 2002). This pper proposes relible step detection method bsed on pttern recognition. From the nlysis of the verticl nd horizontl ccelertion of the foot during one step of the wlking, the signl pttern of wlking behviours is obtined. The stride of the wlker in PNS is scle fctor in ded reckoning lgorithm. Unlike scle fctor of n odometer in cr nvigtion system, the stride in PNS is timevrying prmeter (Mr nd Leu, 1996). The predetermined stride cnnot be used effectively for the distnce mesurement becuse the strides of the wlker re different ccording to the humn prmeters. The stride depends on severl fctors such s wlking velocity, step frequency nd height of wlker etc. As the stride is not constnt nd cn chnge with speed, the step length prmeter must be determined continuously during the wlk to increse the precision. It is suggested tht the stride could be estimted online bsed on liner reltionship between the mesured step frequency nd the stride (Levi nd Judd, 1996). A rel-time step clibrtion lgorithm using Klmn filter with GPS positioning mesurement ws lso proposed (Jirwimut et l,2003). In this pper, we nlyse reltionship between stride, step period nd ccelertion to obtin simple nd robust method of stride determintion. A rel time online estimtion is possible by using only 1-ixis ccelerometer. The combintion of gyroscope nd mgnetic compss hs lredy been pplied in cr nvigtion (Mr nd Leu, 1996) nd it might be very useful heding sensor for pedestrin nvigtion system. However low cost sensor hs importnt drwbcks: A low cost gyro hs lrge bis nd drift error. The mgnetic disturbnces cn be induced ftl compss error. Moreover the error is occurred by n oscilltion of humn body in wlking behviour. In this pper, gyro nd mgnetic compss re integrted using Klmn filter for relible heding of pedestrin. To evlute the performnce of the proposed methods, ctul wlking test in the indoor environment is conducted. The equipment of wlking test is implemented using 1-ixs ccelerometer, 1-xis gyroscope nd mgnetic compss. It consists of two prts: sensor module nd nvigtion computer module. The sensor module is ttched on the nkle. The step number nd stride is computed using the output of the ccelerometer on the sensor module. And wlking direction is obtined from the gyro nd mgnetic compss module. The experiments show the very promising results: less thn 1% step detection error, less thn 5% trvelled distnce error nd less thn 5% heding error. 2 Step detection 2.1 Step behviour nlysis of pedestrin A cycle of humn wlking is composed of two phses: stnding nd wlking phse. The step detection mens recognition of wlking phse. The wlking phse is divided into swing phse nd heel-touch-down phse. Ech phse is shown in figure 1. phse 1st phse 2nd phse Heel- touch-down phse Ground Fig. 1 A wlking behviour In the 1 st swing phse, the foot is locted on behind of grvity centre of humn body. And the foot is locted on front of grvity centre of humn body in the 2 nd swing phse. The foot ccelerted during swing phse. The ccelertion is composed of verticl nd horizontl components s shown in figure 2, where, h, g mens horizontl ccelertion, verticl ccelertion nd grvity force, respectively. Figure 3 nd 4 show motion of leg in the 1 st swing phse nd the 2 nd swing phse respectively.
Kim et l: A Step, Stride nd Heding Determintion for the Pedestrin Nvigtion System 275 h 2.0g. The pttern of ccelertion signl in figure 5 is obtined from 625 times simultions. g Ground Fig. 2 Leg of wlker V H ( h g) cos sin sin + ( h g) cos h g of Verticl direction ( h g) cos h g V sin h g Fig. 3 1 st phse ( h g)cos sin of Verticl direction H cos cos ( h g) sin ( h g)sin of Horizontl direction cos Fig. 4 2 nd phse ( h g)sin + cos ( h g) sin of Horizontl direction The horizontl direction ccelertion nd verticl direction ccelertion during the swing phse is denoted in eqution 1, where ( is inclintion ngle of the leg t time t. H V = ( h g)sin ( + cos ( = ( h g)cos ( sin ( In mny reserches, step is declred when the mesured H or V is lrger thn the threshold. However since the ( depend on chrcteristics of wlking which is different from ech person, it is hrd to determine the exct vlue of threshold of H or V. The step number is miscounted when wrongly predetermined threshold is pplied. By using the signl pttern of ccelertion, this problem cn be solved. Typicl signl pttern of ccelertion is obtined from the computer simultion. We dopted common ssumptions tht typicl inclintion of leg ws within the limit of 30 degree ~ 50 degree nd, h hve rnge of 0.8 ~ 2.3g nd 0.6 ~ (1) Fig. 5 Pttern of horizontl nd verticl ccelertion signl Figure 5 shows the typicl pttern of ccelertion signl on the swing phse. The ccelertion of horizontl direction hs 1 positive pek nd 1 negtive pek in swing phse while the ccelertion of verticl direction hs 1 negtive pek only. The heel-touch-down phse follows the swing phse. A heel-touch-down is impct motion which hits the ground. In heel-touch-down phse, heel hits the ground t first. And then sole of foot nd toe contct with the ground. When the foot hits the ground, the ground repulses the foot. At this time, impct force cts on the foot. The figure 6 shows the heel-touch-down phse. Verticl Axis Horizontl Axis Impc t force Fig. 6 Heel-touch-down phse Ground Repulsive Power Ground Repulsive Power Impct force Heel-touch-down Heel-touch-down Fig. 7 The typicl pttern of signl in heel-touch-down phse
276 Journl of Globl Positioning Systems Figure 7 shows typicl repulsive nd impct force ptterns during the heel-touch-down phse. By combining the swing phse nd heel-touch-down phse in the figures 5 nd 7, we obtin the signl pttern of one wlking cycle. Figure 8 nd 9 show entire signl pttern of the wlking phse. First Second Heel Touch Down Time 2.2 Step detection method To discriminte one cycle of wlking behviour, the signl pttern of swing phse nd heel-touch-down phse in figure 8 nd 9 is dopted. The ccelerometer mesures the signl which is cused by wlking behviour. The step number is counted when ll three phses (1 st swing, 2 nd swing nd heel-touch-down phse) re detected. This method reduces step misdetection probbility nd increse relibility. Recognizing swing nd heel-touchdown pttern using sequentil multi-threshold gives robust nd relible step detection. Also the method cn reduce misdetection probbility of non-wlking behviour such s sitting, turning, kicking nd jumping etc. The detil detection lgorithm is given in figure 11. START Input ccelertion Fig. 8 Verticl ccelertion signl pttern in wlking phse 1st Upper Threshold First Second Heel Touch Down Time 1st Lower Threshold 2nd Upper Threshold Fig. 9 Horizontl ccelertion signl pttern wlking phse 2nd Lower Threshold It is expected intuitively tht the period of heel-touchdown phse is much shorter thn the period of swing phse. The figure 10 shows rel horizontl ccelertion signl in one step. It coincides with the signl pttern model in figure 9. Step Detection END Fig. 11 Flow chrt of step detection First Second Heel Touch Down 3 Stride determintion Fig. 10 Rel horizontl ccelertion signl Becuse the stride is not constnt vlue nd chnges with speed, the stride prmeter must be determined continuously during the wlk to increse its precision. The stride reltes on wlking speed, wlking frequency nd ccelertion mgnitude. In typicl humn wlking behviour, period of one step becomes shorter, stride becomes lrger nd the verticl impct becomes bigger s the wlking speed increses. The reltion between stride, period of one step nd ccelertion is estblished thru the
Kim et l: A Step, Stride nd Heding Determintion for the Pedestrin Nvigtion System 277 ctul wlking test. Figure 12 show test result of two type strides: 60 cm nd 80 cm stride. 4 Heding determintion The gyroscope nd mgnetic compss is widely used to determine heding. The chrcteristics of two sensors re summrized in Tble 3, where dvntge of one sensor is disdvntge of the other. Tb. 3 Comprison between compss nd gyroscope Advntge Disdvntge Mgnetic compss -bsolute zimuth -long term stble ccurcy unpredictble disturbnces externl Fig. 12 The ccelertion signl of 60 cm nd 80 cm stride The tester wlks with the fixed stride using ground mrks. In the figure, the reltion between the ccelertion nd stride is clerly shown. The tbles 1 nd 2 show reltion between ccelertion nd one step time in this test. The longer stride induces the bigger ccelertion. However difference of one step time is hrd to pply stride determintion becuse of smll difference in mesurements. Tb. 1 The men of ccelertion bsolute vlue Stride Men vlue (g) Gyro scope -no externl disturbnces -short term ccurcy reltive zimuth drift From tble 3, n optiml nd relible system might be expected by integrting the gyroscopes with the mgnetic compss. In the integrted system, the gyroscope cn correct the mgnetic disturbnces, t the sme time the compss cn determine nd compenste the bis of the gyros nd the initil orienttion. The combintion of gyroscope nd mgnetic compss hs lredy been pplied in the cr nvigtion system. The integrtion method of the gyroscope nd the mgnetic compss used in this pper is given in Figure 13. 60cm 0.2882 80cm 0.5549 Gyroscope Gyro bis Angulr rte Heding error Heding Tb. 1 The period of one step Initil Heding Compss rte Disturbnce detection Klmn Filter Stride Men of time (sec.) Mgnetic compss Heding 60cm 0.675 80cm 0.662 Eqution 2 is the experimentl eqution obtined from severl wlking tests, where mens the mesured ccelertion nd represents the number of smple in one cycle of wlking. The eqution represents the reltion between mesured ccelertion nd stride. It is used for online estimtion of the stride. 3 N k = 1 A k Stride( m) = 0.98 (2) N Fig. 13 Scheme for n integrtion of gyroscope nd mgnetic compss When the pedestrin is wlking, the influence of mgnetic disturbnce sources chnges unpredictbly, creting error in the compss heding. This error degrdes the performnce of integrtion system. The impct of error cn be reduced by detecting the disturbnce. The error cn be observed vi the ngulr rte of compss heding: ψ compss ( tk + ψ compss ( tk ) ω compss = (3) t where ω is ngulr rte, ψ is heding nd t is the time intervl. The disturbnce cn be detected when difference of compss ngulr rte ω compss nd gyroscope ngulr rte ω gyro is lrger thn given threshold. The compss mesurement is ignored. The sttes of Klmn filter re heding error nd sensor error (gyro bis).
278 Journl of Globl Positioning Systems 5 Experiments In order to evlute the performnce of the proposed method, the ctul wlking test is done. The tester is mle ged 26 with 175cm height. The experiments re done t the 4th floor hllwy of the engineering building, Chungnm Ntionl University, Dejeon, Kore. In the experiments, wlking distnce determintion nd heding determintion re crried out seprtely. 5.1 Experimentl setup Figure 14 shows the experimentl equipments. 16- bit Microcontroller MEMORY POWER Compss Communiction Nvigtion Computer Bluetooth Dt Acquisition Fig. 15 The output signl of ccelerometer Sensor Module Accelermeter Gyro Fig. 14 Experimentl equipment The experimentl equipments consist of the sensing module, the nvigtion computer nd dt cquisition system (notebook computer). The body-worn sensing module consists of 16-bit microcontroller, MEMS ccelerometer (ADXL105, Anlog device Inc.), gyroscope (MEMS DMU, Crossbow Inc.), low-cost digitl mgnetic compss sensor (CMPS03, ROBOT Electronics Inc.) nd other electricl prts (RS-232 converter, DC-DC converter, 9V bttery, Bluetooth modem). The sensor module is ttched on the nkle with horizontl direction s shown figure 14. Fig. 16 Estimted stride in 1 st test 5.2 Experiment of wlking distnce determintion To evlute performnce of the step detection nd stride determintion lgorithm, the tester ws sked to wlk for pre-determined pth (74.2m nd 145.6m stright pth). Figure 15 shows the output of ccelerometer. The true step number of first test is 100 steps nd second test 200 steps. The stride is determined using eqution 2. The figure 16 nd 17 show the strides of left leg. The men of estimted stride is obtined s 76.1 cm nd 75.9 cm respectively. Tble 4 shows result of wlking test in detil. Fig. 17 Estimted stride in 2 nd test In the 1st test, the proposed method count step number without loss, while the 2 step detection is lost in 2nd test. The 2 step loss is hppened in the lst 199th nd 200th step where the wlking pttern is bruptly chnging. The wlking distnce error is obtined 2.5m, 6.1m respectively. The trvelled distnce with less thn 5% error is obtined. These results verify tht the proposed method cn mesure ccurte step numbers nd distnce.
Kim et l: A Step, Stride nd Heding Determintion for the Pedestrin Nvigtion System 279 1 st test 2 nd test Tb. 4 The mesured wlking distnce Actul wlking behvior Step number Wlking distnce Step number Wlking distnce 100 step 74.2m 200 step 145.6m Mesured step number Mesured wlking distnce 100 step 76.728 m 198 step 151.674 m detection, we nlyse the verticl nd horizontl ccelertion of the foot during one step of the wlking. With this nlysis, new step determintion bsed on the pttern recognition is proposed nd the step number cn be counted ccurtely. The reltionship between stride nd ccelertion is derived from ctul test. An efficient stride determintion method where the stride cn be estimted online, so tht the user does not need to specify his/her stride, is proposed. The integrtion scheme of the gyro nd mgnetic compss is proposed for error compenstion of gyro nd disturbnce rejection of mgnetic compss. The experiments using the ctul wlking tests in indoor shows tht the proposed method gives less thn 1% step, 5% trvelled distnce nd 5% heding errors. It is expected tht the proposed PNS will be very useful nvigtion system for pedestrin nvigtion. 5.3 Experiment of heding determintion For heding determintion test, the tester wlks stright pth of north direction. Figure 18 is result of heding determintion test. Fig. 18 Estimted heding by gyro nd by integrtion In the figure, the heding of stnd-lone gyro shows oscilltory errors due to the body motion. Klmn filter in the integrted system reduces these errors. The experiments show tht the heding of pedestrin cn be determined with ccurcy of 5 degree. 6 Conclusions nd outlook This pper proposes methods to estimte the PNS DR prmeters: step, stride nd heding. For ccurte step References Gbglio, V., (2003): GPS/INS Integrtion for Pedestrin Nvigtion Ph. D. disserttion. Institute of Geomtics of the Swiss Federl Institute of Technologye in Lusnne. Mr, J., (1996): Simultions of the positioning ccurcy of integrted vehiculr nvigtion systems. In: J.-H. Leu(Eds.): Proc. Inst. Elect. Eng. Rdr, Sonr Nvigtion, vol. 143, Apr., 121 128. Ldetto, Q., (2002): In Step with INS. In: B. Merminod (Eds.):GPS WORLD mgzine, 30-38 Quentin Ldetto (2000): On foot nvigtion: continuous step clibrtion using both complementry recursive prediction nd dptive Klmn filtering. Proceedings of ION GPS 2000, 1735~1740. Jirwimut, R., (2003): A Method for Ded Reckoning Prmeter Correction in Pedestrin Nvigtion System. In: P. Ptsinski; V. Grj; F. Cecelj; W.Blchndrn (Eds.):IEEE Trnsctions on Instrumenttion nd Mesurement, Vol. 52,.1. Levi, R. W., (1996): Ded Reckoning Nvigtionl System using Accelerometer to Mesure Foot Impcts. In: T. Judd,(Eds.): United Sttes Ptent. 5,583,776 Lee, S.-W., (2001): Recognition of Wlking Behviours for Pedestrin Nvigtion. In: K. Mse,(Eds.): Proc. 2001 IEEE Int l Conf. Control Applictions (CCA 01), IEEE Control Systems Soc., Pisctwy, N.J., 1152 1155. Gbglio, V., (1999): Rel-time clibrtion of length of steps with GPS nd ccelerometers. In: B. Merminod (Eds.):Proceeding of GNSS 99, 599 605.