A Battery-free Object Localization and Motion Sensing

A Battery-free Object Localization and Motion Sensing

Platform

Yi Zhao

EE Dept. University of Washington yizhao@https://www.360docs.net/doc/9013788536.html,

Anthony LaMarca

Intel Research

https://www.360docs.net/doc/9013788536.html,marca@https://www.360docs.net/doc/9013788536.html,

Joshua R.Smith

CSE&EE Dept.

University of Washington

jrs@https://www.360docs.net/doc/9013788536.html,

ABSTRACT

Indoor object localization can enable many ubicomp applica-tions,such as asset tracking and object-related activity recog-nition.Most location and tracking systems rely on either battery-powered devices which create cost and maintenance issues or cameras which have accuracy and privacy issues. This paper introduces a system that is able to detect the3D position and motion of a battery-free RFID tag embedded with an ultrasound detector and an https://www.360docs.net/doc/9013788536.html,bining tags’acceleration with location improves the system’s power management and supports activity recognition.We charac-terize the system’s localization performance in open space as well as in a cluttered wet lab environment.The system is used to track real-time location and motion of the tags as well as recognize actions performed on the objects to which the tag is attached.The median localization accuracy is7.6cm–(3.1,5,1.9)cm for each(x,y,z)axis–with max update rates of15Sample/s using single RFID reader antenna.

Author Keywords

RFID;localization;Object tracking;Activity Recognition ACM Classi?cation Keywords

C.2.4Distributed Systems:Distributed Applications

1.INTRODUCTION

Accurate3D object location is one of the key context infor-mation enabling many smart home and smart of?ce capabil-ities,such as activity inference,asset tracking,and human-robot interaction.Unfortunately,most localization systems are based on either cameras[10]or battery-powered dis-tributed sensing devices[4,6].Camera-based systems cre-ate a host of privacy issues and have trouble localizing trans-parent,re?ective and irregular-shape objects.The latter sys-tems typically have better accuracy and identi?cation ability, but don’t scale well due to tag size,battery life and mone-tary and maintenance costs of the active sensors,especially in dense localization sensor network.RF wireless power technology[3,5,11]can power varied sensing devices by available RF infrastructure rather than battery,such as pas-sive RFID tags.Those RFID tags are small,cheap and have Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for pro?t or commercial advantage and that copies bear this notice and the full cita-tion on the?rst page.Copyrights for components of this work owned by others than ACM must be honored.Abstracting with credit is permitted.To copy otherwise,or re-publish,to post on servers or to redistribute to lists,requires prior speci?c permission and/or a fee.Request permissions from Permissions@https://www.360docs.net/doc/9013788536.html,.

Ubicomp′14,September13-172014,Seattle,WA,USA

Copyright c 2014ACM978-1-4503-2968-2/14/09˙..$15.00

https://www.360docs.net/doc/9013788536.html,/10.1145/2632048.2632078long life-time due to its battery-free,thereby having potential as low-cost localization system.Besides,the RFID proto-cols are optimized for reading dense tag population,therefore allowing good scaling.Current RFID localization systems can provide at best sub-meter level coarse localization using RF signal strength.However,researchers have to deal with the trade-off between low accuracy due to RF multi-path and high-cost because of using large number of RFID reference nodes[2].

To address these issues,this paper introduces the?rst battery-free acoustic localization platform based on RFID infras-tructure.In previous work,we demonstrated that a passive RFID tag augmented with an ultrasonic microphone can ac-curately estimate its range from an ultrasonic transmitter[11]. In this work,both our ultrasonic transmitters and receivers were built-into fully-programmable passive tags(Figure1) named WISPs(Wireless Identi?cation Battery-free Sensing Platform[9]),which were augmented with either ultrasound beacons or ultrasound microphones and accelerometers.By having all of the ultrasonic sensing and actuation in the tag, we can make use of commercial RFID readers to power and communicate with the tags.In this paper,we show that this basic acoustic ranging capability can be extended to robust, accurate3D localization in a lab environment.We do this by having tags perform ultrasound time of?ight(ToF)mea-surements to estimate its range to four ultrasonic transmitters in known locations.These ranging estimates infer the tag’s 3D location relative to the transmitters by applying a translit-eration method that can mitigate the effect of indirect-path and multi-path acoustic re?ections.The result is a scalable, battery-free3D localization system with the following fea-tures:

?7.6cm accuracy overall with(3.1,5,1.9)cm accuracy for each(x,y,z)axis for line of sight stationary tags.

?Average update rate of153D location estimates per second (given1W RF power and1.8m test range).

?Maximum range of3.5m(at best RF sensitive

direction). Figure1.A battery-free ultrasonic localization tag that is powered and read by RF signals.

Figure2.Sensing and communication protocol:in the blue area,the Sense WISP is awake;in the white region,it asleep.

?252μJ Energy consumption for positioned tag updating one location measurement.

?Tolerance to ultrasound echo noise.

?Wireless powered acoustic transmitters that allow more freedom for system set-up.

?Accelerometer data which accompanies location estimates. This enables both motion and activity estimation as well as improves tag’s power management

2SYSTEM ARCHITECTURE

Our system is designed to allow an customized RFID tag to detect the time of arrival(ToA)of ultrasound sending from four ultrasound transmitters,and to measure its3-axis accel-eration data.The tag will then send back these results back to one RFID reader in range using its Electronic Product Code (EPC)(tag’s ID).Finally,the RFID reader will report all tag readings to a client to compute the tag-to-transmitter distance ranges,and from that estimate the tag’s?nal3D location and motion state.To simplify the explanation,we use a basic two tags system to demonstrate how to localize a target tag us-ing single RFID tag,antenna and reader.The basic system consists of three hardware parts as follows and the design of hardware can be found in previous work[11]:

2.1Sense WISP

The Sense WISP is the target tag to be positioned,which is a battery-free programmable RFID tag(Figure1)augmented with an ultrasound detector and3D accelerometer.This small tag is compatible with RFID UHF EPC Gen2protocol.We are intended to use commercial RFID reader and another ref-erence RFID tag(called Spy WISP)to estimates the3D lo-cation of the Sense WISP.The Spy WISP sent ultrasound from its four ultrasonic beacons.All RFID tag mentioned in this work is wireless powered by1W RF signals sent from single nearby RFID reader antenna.During the step of the RFID protocol called the”Inventory Round”,the Sense WISP starts a timer in anticipation of the?rst ultrasound pulse be-ing simultaneously sent from the Spy WISP and then mea-sures the Time of arrival(ToA)of all four sent ultrasound signals.Time-division multiplexing is used to schedule4ul-trasound transmission and detection within one RFID inven-tory round.To avoid interference between ultrasound signals from different emitters in different RFID reading rounds,the Spy WISP con?nes all of its ultrasound signals to a10ms window.This is the minimum“inventory period”needed by the RFID reader to update one Sense WISP EPC read-ing.The intervals between ultrasound pulses are chosen to be long enough that ultrasound collisions cannot occur.After ultrasound measurement,the Sense WISP samples its3D ac-celeration and send all those data back to the reader.The data being sent encodes the RFID readings along with tag’s unique ID(that is the Electronic Product Code(EPC))(Figure2).The accelerometer in the Sense WISP has built-in activity and in-activity detection circuitry and can measure up to±2G with 16mG sensitivity.

2.2Spy WISP

The Spy WISP RFID tag is in many ways like the Sense WISP tag:It is also a passive RFID tag and can understand and follow the EPC protocol.Unlike the Sense WISP,how-ever,the Spy WISP tag does not acknowledge any radio re-sponses to the RFID reader,nor does perform any sensing.In-stead,the Spy WISP is equipped with425kHZ narrow band ultrasound beacons at known location.The Spy WISP’s func-tion is to transmit an ultrasonic pulse from each beacons at a time known to the Sense WISP(Figure2).The Spy WISP synchronizes the Sense WISP by listening a hard-coded in-formation in RFID command sent from reader to Sense WISP in each tag inventory round.The details of synchronization method can be found in previous paper[11].In a brief,the ToA measurements are achieved using the EPC protocol it-self:Both the Sense WISP and Spy WISP listen for the start of the”Inventory Round”of the EPC protocol.On observ-ing the signal,the Spy WISP emits a sequence of ultrasound pulses at known intervals.The Sense WISP using this sig-nal to start a time and being listening for the ultrasonic chips. Since the EPC protocol signals are sent via radio,both tags will receive and recognize the signal at the nearly the same time and from the perspective of acoustic wave speeds,any difference is negligible.Since the Spy WISP does not send any radio replies,it does not interfere with any tags replying to the reader’s read requests.As it turns out,the transmission of ultrasound makes the Spy WISP more power hungry than Sense WISP,it thus it has to be placed within1m of the RFID antenna.The result,however,is an extremely?exible system with mobile,battery-free tags along with a small number of battery and wire-free reference tags to be placed in the envi-ronment,independent from the reader.

2.3RFID Reader and Client Application

We use an Impinj EPC Gen2UHF reader to power and com-municate with the Sense WISP by issuing inventory com-mands.By sending valid inventory commands to the Sense WISP,the reader helps to triggers the acoustic transmission from transmitters in the Spy WISP and detection of the ToA in Sense WISP.In response,the Sense WISP will passively re-?ect back to reader its EPC information(include Sense WISP

ID,4ToA measurement results,3axis acceleration data,mo-tion activity state from previous inventory round)(Figure2), which is used for inferring tag’s location and motion state. SOFTWARE DESIGN

3.13D Location Calibration

In general,the time of arrival of ultrasound signals in a given environment is linearly proportional to the distance it prop-agated.However,the hardware and sensor introduces some synchronization error.Therefore,we use a?rst order poly-nomial line?t model to matching Sense WISP-to-transmitter

distance d T

i and its ToA measurement result T R

i

.The line?t

model can be expressed as d T i=a i×T R

i +b i(i=1,2,3).

Using this model,the small RF synchronization offset,Sense WISP detection clock offset can be modeled in b i,and the speed of sound variance can be modeled in a i.After that,we can calculate the position of Sense WISP with these calcu-lated distances d T i and transmitter position(x i,y i,z i)using a trilateration algorithm[7].Besides,a time and tag’s motion weighted Kalman?lter is used to?lter out measurement and environmental noise.A python based RFID localization ap-plication GUI is developed to visualize real time tag tracking result.(The GUI uses a python reader client packages from [8])

3.2Power Optimization

The?rmware and hardware of the Sense and Spy WISP are optimized for low power and duty cycled during RFID com-munication and ultrasound detection(Figure2).The Sense WISP only consumes about80μW during a wake up cycle and10μW for a sleep cycle.As a result,the system is able to more accurately track static or slow-moving tags than ones moving quickly.As a further optimization,stationary tags don’t need to perform their location estimation continually since they are not moving.We use the accelerometer data on Sense WISP to implement this power-savings.Once the tag is detected to be stationary,a Sense WISP will increase its sleep cycle length,once motion is resumed,will increase the localization update rate to track its new position.

4EXPERIMENT AND PERFORMANCE

We test the system with single Spy WISP and single Sense WISP to be located in a1.2m?1.2m?1.8m open space and test with more Sense WISPs in a Wet Lab later.The Spy WISP’s four ultrasound transmitters are placed at the4ver-tices of15cm-long square with known location which are in line of sight for the Sense WISP.We use only one reader pro-duces1W power with single antenna(6dBi gain),which can power up Sense WISP up to3.5m away in best sensitive direc-tion.The total energy used for Sense WISP updating one3D location is below252μJ.The experimental results are shown in Figures3and4.In theory,using the RFID synchronization and acoustic TDMA method[11],the system can support lo-calizing up to65535Sense WISPs,however,the localization update rate for each tag will relatively decrease if the system scales up.

Accuracy:Because of the geometry setup of the ultrasounds transmitters,the system can obtain better accuracy in z

axis,Figure3.Free space localization performance:CDF of localization error distribution,in percentage and absolute units.

which is perpendicular to the planes of transmitters.3D ge-ometry setup of those transmitters can improve the accuracy for other axis,however,the ground truth measurement for that is challenging for experiment.Figure4shows that ac-curacy decreases as the Sense WISP-to-transmitters distance or ultrasound angle of arrival increase.Because the attenu-ation of ultrasound and directional ultrasound sensor intro-duces non-liner noise which can not be modeled well using line-?tting model at those circumstance.The average accu-racy is(3.1,5,1.9)cm for each(x,y,z)direction within test-ing range.

Range:The system maximum localization range is3.4m (when x=0,y=0,z=3.4m).The range of the system is limited by the amount of available power for Sense WISP, that means when the Sense WISP is out of the range,it is not able to powered up or complete one detection.

Update Rate:The update rate for this system is limited by the available RF power for Sense WISP,its energy harvest-ing ef?ciency and RFID protocol(maximum45S/s for Sense WISP),and is5S/s31S/s for this experiment.It will de-crease as the population of Sense WISP increase.The average update rate is15S/s and can be adjusted by changing the sleep time of either Spy WISP and Sense WISP.Spreading high-gain omni-directional antenna can offer better power density in the space,thereby improve the update rate if needed. Acceleration detection:the system is able to detect the move-ment of Sense WISP tagged object.By grasping the tagged objects from table,the system have90%con?dence of de-tecting the movement for15test samples within1m tag-to-reader range.In order to capture those fast movement of tag, like being grasped,the reader have to read Sense Tag with a relatively higher rate than the movement,which is limited by RFID protocol and available RF power near the Sense WISP. For more static tags whose orientation is known to the sys-tem,our system can detect the change of its orientation even at a lower Sense WISP update rate up to1.8m range using single antenna.More antennas in test space will also help. Therefore the platform have distinguished motion detection ability varied by available RF power in the space. APPLICATION AND IMPLEMENTATION

One high-value application of context-awareness(and one of the largest challenges facing science)is the documentation and replication of scienti?c lab experiments.A2012study, for example,showed that47of53key cancer biology results

Figure 4.Localization accuracy in 3D open space:sub-?gures 1?3show results at z =94cm,z =134,z =174cm .The 4ultrasound sources in are placed at the vertices of the 16cm ×14cm cyan squre in the center of the chess plane (20cm/grid,origin at the center).The yellow ball is the ground truth location with sensor’s acutal size (radius =8mm ).The blue ball shows the localization result;size represents standard deviation on each

axis.

Figure 5.Wetlab.We are exploring the use of the localization tags for activity inference in a wetlab

scenario.

Figure 6.Wetlab localization performance:CDF of localization error distribution,in percentage and absolute units.The performance is not signi?cantly different than the freespace case.

could not be replicated,in some cases even with the help of the original researchers [1].Part of what makes documenta-tion and replication dif?cult is the large number and subtlety the many steps performed during a lab experiment.Another factor is that often documentation is performed after the fact since both hands are needed to perform experiment,and lab technicians often wear protective gloves that make writing or typing dif?cult.

Accordingly,we are using a self-documenting smart wet lab as a guiding application of our location system,using a clut-tered lab bench as our deployment environment (Figure 5).We are RFID-tagging key objects in the lab (e.g.:bottles of reagents and glass containners)and tracking their location and motion as experiments https://www.360docs.net/doc/9013788536.html,ing accurate,high-update rate location readings from our tags,we aim to not only infer which speci?c objects are present on the bench,but

when,where and how they are used.With this detailed object location and motion data,coupled with representations of the experimental protocols we hope to be able to infer and docu-ment the objects being used,actions performed on/with them and recognize the experiment being performed.

We implement this system in an (0.8m ×0.5m ×1.2m )lab bench (one table and one shelf above)in the smart wet lab (Figure 5)and attach Sense WISP tag to several bottles of reagents,the system can then infer the current state and ac-tivity of those bottles by measuring its location and move-ment(such as it is on the shelf or in media preparing area)and infer their activity and state.An experiment is designed to report the current static location of 1object,2objects or three objects in 14different locations.The average accuracy can be seen from Figure 6.

Another experiment is conducted to try to recognize a pouring event,when the Sense WISP tagged bottle is pouring into an-other container in order to alarm people if the reagents in the tagged bottle is pouring into the wrong container.Two iden-tical container with the same distance to the tagged bottle are used as the destination of the pouring.During the test,the researcher grasps the tagged bottle from the table and pour-ing its liquid into one of the two containers,which location is known,once any movement activity of the bottle is detected and the bottles orientation is turning to horizontal,the system computes the current location of the bottle while it is pouring.The system then can estimate which container the tagged bot-tle is pouring into by ?nd which container it is mostly close to.Placing the pair of container in 5different places,the system can detect the correct pouring destination container with 90%con?dence for 8times the pouring event for each location.

CONCLUSION AND FUTURE WORK

Using simple RFID infrastructure setting for wireless power and communication,our system is able to estimate the 3D position of a battery-free ultrasound sensor tag with 8cm ac-curacy in an acoustically complex indoor environment.The same tags also provide acceleration sensing.The system is designed to support activity recognition and tracking,by jointly interpreting the data from these sensors.Future work will improve localization performance,scale up the number of sensors with multiple antennas,and test the system’s effec-tiveness in activity recognition applications.

ACKNOWLEDGMENT

This work is funded by Intel Science and Technology Center for Pervasive Computing(ISTC-PC)and NSF award number CNS1305072.REFERENCES

1.Begley,C.,and Ellis,L.Drug development:Raise

standards for preclinical cancer research.Nature483

(2012),531533.

2.Bouet,M.,and Dos Santos,A.L.R?d tags:Positioning

principles and localization techniques.In Wireless Days, 2008.WD’08.1st IFIP,IEEE(2008),1–5.

3.Carter,M.,Stetter,J.,Smith,J.,Parks,A.,Zhao,Y.,

Findlay,M.,and Patel,V.Printed low power

amperometric gas sensors employing rf energy

harvesting.In Meeting Abstracts,no.44,The

Electrochemical Society(2012),1602–1602.

4.Ni,L.M.,Liu,Y.,Lau,Y.C.,and Patil,https://www.360docs.net/doc/9013788536.html,ndmarc:

indoor location sensing using active r?d.Wireless

networks10,6(2004),701–710.

5.Parks,A.N.,Sample,A.P.,Zhao,Y.,and Smith,J.R.A

wireless sensing platform utilizing ambient rf energy.In Biomedical Wireless Technologies,Networks,and

Sensing Systems(BioWireleSS),2013IEEE Topical

Conference on,IEEE(2013),154–156.

6.Priyantha,N.B.,Chakraborty,A.,and Balakrishnan,H.

The cricket location-support system.In Proceedings of the6th annual international conference on Mobile

computing and networking,ACM(2000),32–43.

7.Rahman,M.Z.,and Kleeman,L.Paired measurement

localization:a robust approach for wireless localization.

Mobile Computing,IEEE Transactions on8,8(2009),

1087–1102.

8.Ransford,B.Python client for LLRP-based RFID

readers.https://https://www.360docs.net/doc/9013788536.html,/ransford.

9.Sample,A.P.,Yeager,D.J.,Powledge,P.S.,Mamishev,

A.V.,and Smith,J.R.Design of an r?d-based

battery-free programmable sensing platform.

Instrumentation and Measurement,IEEE Transactions on57,11(2008),2608–2615.

10.Yilmaz,A.,Javed,O.,and Shah,M.Object tracking:A

survey.Acm computing surveys(CSUR)38,4(2006),

13.

11.Zhao,Y.,and Smith,J.R.A battery-free r?d-based

indoor acoustic localization platform.In RFID(RFID), 2013IEEE International Conference on,IEEE(2013),

110–117.