This article has been accepted for inclusion in a future issue of this journal.Content is final as presented,with the exception of pagination. XIE eraL:MULTI-TOUCH IN THE AIR:CONCURRENT MICROMOVEMENT RECOGNITION USING RF SIGNALS 035 03 025 0. 0.15 0.0 SL SR RR RL ZI speed Medium Speed Low AB BC AC ABO (a) (b) (e) (d) Fig.16.Micromovement recognition latency and accuracy results.(a)Latency for different recognition schemes.(b)Accuracy for different moving speeds. (c)Accuracy for different orientation deviations.(d)Accuracy with different number of antennas. 1)Robustness to Various Moving Speeds:Phase Profiles is 85%,whereas the accuracy with antenna B is 66%,as the achieves fairly good performance in recognition accuracy, micromovement is performed more close to the central beam with an average accuracy of 77%in high speed mode and of antenna A than B on average. 94%in low speed mode,respectively.We set the duration of micromovements to three different levels,i.e.,high speed VI.CONCLUSION (0.5s),medium speed(1s)and low speed (3~4s).Fig.16(b) In this paper,we propose RF-Glove,an RF based motion plots the recognition accuracy for different moving speeds. recognition system that can recognize multiple finger-based We find that the accuracy decreases in high speed,since micromovement.It uses one COTS RFID reader with 3 our template phase profiles are extracted from the samples antennas and five passive tags attached to the five fingers. mostly with medium speed.The performance degradation is We propose a phase profiling based approach to RF signal mainly caused by the difference in sampling granularity and based finger micromovement recognition.We implemented micromovement distortion due to the high speeds.Moreover, RF-Glove using COTS RFID systems and evaluated its perfor- the recognition accuracy slightly increases in low speed mode, mance in realistic settings.The experiment results show that as the finer sampling granularity helps improve the recognition we achieve an average accuracy of 92.1%where the number accuracy. of finger micromovement types is eight. 2)Robustness to Various Orientation Deviations:Phase Profiles achieves fairly good performance in recognition accu- REFERENCES racy,with an average accuracy of 80%in small deviation mode [1]Kinect.(2016).[Online].Available:http://www.microsoft.com/en-us/ and 60%in large deviation mode.respectively.We evaluated kinectforwindows [2]Wii.(2016).[Online].Available:http://wii.com the impact of orientation deviations of micromovement on the [3]Leap Motion.(2016).[Online].Available:http://www.leapmotion.com performance in accuracy.We set the orientation deviation of [4]J.Han et al.,"GenePrint:Generic and accurate physical-layer identi hands to no deviation(0°)，small deviations(15w20°)and fication for UHF RFID tags."IEEE/ACM Trans.Nerw:.vol.24,no.2. large deviations (30~40).Fig.16(c)plots the recognition Pp.846-858,Apr.2016. [5]Y.Zheng and M.Li."P-MTI:Physical-layer missing tag identifica- accuracy for different orientation deviations.We observe that tion via compressive sensing,"in Proc.IEEE INFOCOM,Apr.2013. the accuracy more or less decreases in both situation of Pp.917-925. [6]J.Ou,M.Li,and Y.Zheng,"Come and be served:Parallel decoding small deviations and large deviations,since our template for COTS RFID tags,"in Proc.ACM MobiCom.2015.pp.500-511. phase profiles are extracted from the samples mostly with no [7]M.Chen,W.Luo,Z.Mo,S.Chen,and Y.Fang."An efficient tag search obvious deviations.The deviations of the test phase profiles protocol in large-scale RFID systems with noisy channel,"IEEEACM greatly reduce their similarity with the actual reference phase Trans.Nerw..vol.24.no.2,pp.703-716.Apr.2016. [8】J.Liu et al.,“"Efficient RFID grouping protocols,”IEEE/ACM Trans.. profiles.Nevertheless,they all achieve recognition accuracy of Nene,vol.24,no.5,Pp.3177-3190,0ct2016. at least 60%. [9]X.Liu,B.Xiao,S.Zhang,and K.Bu,"Unknown tag identification in 3)Robustness to Different Number of Antennas:Phase Pro- large RFID systems:An efficient and complete solution,"IEEE Trans. Parallel Distrib.Syst.,vol.26,no.6,pp.1775-1788,Jun.2016. files achieves fairly good performance in recognition accuracy, [10]J.Wang.D.Vasisht,and D.Katabi,"RF-IDraw:Virtual touch screen in with an average accuracy of at least 66%with different the air using RF signals,"in Proc.ACM S/GCOMM,2014,pp.235-246. [11]J.Wang and D.Katabi,"Dude,where's my card?:RFID positioning that numbers of antennas.We set the number of antennas from works with multipath and non-line of sight,"in Proc.ACM S/GCOMM. 1 to 3,and let the volunteer perform the micromovement at 2013,pp.51-62. the position under antenna A.For fairness,we only compare [12]L.Yang et al."Tagoram:Real-time tracking of mobile RFID tags to high precision using COTS devices,"in Proc.ACM MobiCom,2014. the recognition performance on the five small-range micro- Pp.237-248. movements,since the large-range movements require three [13]L.Shangguan.Z.Li,Z.Yang.M.Li,and Y.Liu."OTrack:Order track- antennas to perform 3D positioning.Fig.16(d)plots the ing for luggage in mobile RFID systems,"in Proc.IEEE INFOCOM. accuracy for different number of antennas.We find that as Apr.2013,Pp.3066-3074. [14]L.M.Ni,Y.Liu,Y.C.Lau,and A.P.Patil,"LANDMARC:Indoor the number of antennas decreases,the average recognition location sensing using active RFID,"Wireless Nenw.,vol.10,no.6. accuracy gradually decreases from 92.1%to 66%.Moreover, Pp.701-710.2004. the recognition accuracy for different antenna varies even if the [15]J.Wang,F.Adib,R.Knepper,D.Katabi,and D.Rus,"Rf-compass: Robot object manipulation using RFIDs,"in Proc.ACM MobiCom,2013, number of antennas is equal,e.g.,the accuracy with antenna A Pp.3-14This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. XIE et al.: MULTI-TOUCH IN THE AIR: CONCURRENT MICROMOVEMENT RECOGNITION USING RF SIGNALS 13 Fig. 16. Micromovement recognition latency and accuracy results. (a) Latency for different recognition schemes. (b) Accuracy for different moving speeds. (c) Accuracy for different orientation deviations. (d) Accuracy with different number of antennas. 1) Robustness to Various Moving Speeds: Phase Profiles achieves fairly good performance in recognition accuracy, with an average accuracy of 77% in high speed mode and 94% in low speed mode, respectively. We set the duration of micromovements to three different levels, i.e., high speed (0.5s), medium speed (1s) and low speed (3∼4s). Fig. 16(b) plots the recognition accuracy for different moving speeds. We find that the accuracy decreases in high speed, since our template phase profiles are extracted from the samples mostly with medium speed. The performance degradation is mainly caused by the difference in sampling granularity and micromovement distortion due to the high speeds. Moreover, the recognition accuracy slightly increases in low speed mode, as the finer sampling granularity helps improve the recognition accuracy. 2) Robustness to Various Orientation Deviations: Phase Profiles achieves fairly good performance in recognition accu￾racy, with an average accuracy of 80% in small deviation mode and 60% in large deviation mode, respectively. We evaluated the impact of orientation deviations of micromovement on the performance in accuracy. We set the orientation deviation of hands to no deviation (0◦), small deviations (15 ∼ 20◦) and large deviations (30 ∼ 40◦). Fig. 16(c) plots the recognition accuracy for different orientation deviations. We observe that the accuracy more or less decreases in both situation of small deviations and large deviations, since our template phase profiles are extracted from the samples mostly with no obvious deviations. The deviations of the test phase profiles greatly reduce their similarity with the actual reference phase profiles. Nevertheless, they all achieve recognition accuracy of at least 60%. 3) Robustness to Different Number of Antennas: Phase Pro- files achieves fairly good performance in recognition accuracy, with an average accuracy of at least 66% with different numbers of antennas. We set the number of antennas from 1 to 3, and let the volunteer perform the micromovement at the position under antenna A. For fairness, we only compare the recognition performance on the five small-range micro￾movements, since the large-range movements require three antennas to perform 3D positioning. Fig. 16(d) plots the accuracy for different number of antennas. We find that as the number of antennas decreases, the average recognition accuracy gradually decreases from 92.1% to 66%. Moreover, the recognition accuracy for different antenna varies even if the number of antennas is equal, e.g., the accuracy with antenna A is 85%, whereas the accuracy with antenna B is 66%, as the micromovement is performed more close to the central beam of antenna A than B on average. VI. CONCLUSION In this paper, we propose RF-Glove, an RF based motion recognition system that can recognize multiple finger-based micromovement. It uses one COTS RFID reader with 3 antennas and five passive tags attached to the five fingers. We propose a phase profiling based approach to RF signal based finger micromovement recognition. We implemented RF-Glove using COTS RFID systems and evaluated its perfor￾mance in realistic settings. The experiment results show that we achieve an average accuracy of 92.1% where the number of finger micromovement types is eight. REFERENCES [1] Kinect. (2016). [Online]. Available: http://www.microsoft.com/en-us/ kinectforwindows [2] Wii. (2016). [Online]. Available: http://wii.com [3] Leap Motion. (2016). [Online]. Available: http://www.leapmotion.com [4] J. Han et al., “GenePrint: Generic and accurate physical-layer identi- fication for UHF RFID tags,” IEEE/ACM Trans. Netw., vol. 24, no. 2, pp. 846–858, Apr. 2016. [5] Y. Zheng and M. Li, “P-MTI: Physical-layer missing tag identifica￾tion via compressive sensing,” in Proc. IEEE INFOCOM, Apr. 2013, pp. 917–925. [6] J. Ou, M. Li, and Y. Zheng, “Come and be served: Parallel decoding for COTS RFID tags,” in Proc. ACM MobiCom, 2015, pp. 500–511. [7] M. Chen, W. Luo, Z. Mo, S. Chen, and Y. Fang, “An efficient tag search protocol in large-scale RFID systems with noisy channel,” IEEE/ACM Trans. Netw., vol. 24, no. 2, pp. 703–716, Apr. 2016. [8] J. Liu et al., “Efficient RFID grouping protocols,” IEEE/ACM Trans. Netw., vol. 24, no. 5, pp. 3177–3190, Oct. 2016. [9] X. Liu, B. Xiao, S. Zhang, and K. Bu, “Unknown tag identification in large RFID systems: An efficient and complete solution,” IEEE Trans. Parallel Distrib. Syst., vol. 26, no. 6, pp. 1775–1788, Jun. 2016. [10] J. Wang, D. Vasisht, and D. Katabi, “RF-IDraw: Virtual touch screen in the air using RF signals,” in Proc. ACM SIGCOMM, 2014, pp. 235–246. [11] J. Wang and D. Katabi, “Dude, where’s my card?: RFID positioning that works with multipath and non-line of sight,” in Proc. ACM SIGCOMM, 2013, pp. 51–62. [12] L. Yang et al., “Tagoram: Real-time tracking of mobile RFID tags to high precision using COTS devices,” in Proc. ACM MobiCom, 2014, pp. 237–248. [13] L. Shangguan, Z. Li, Z. Yang, M. Li, and Y. Liu, “OTrack: Order track￾ing for luggage in mobile RFID systems,” in Proc. IEEE INFOCOM, Apr. 2013, pp. 3066–3074. [14] L. M. Ni, Y. Liu, Y. C. Lau, and A. P. Patil, “LANDMARC: Indoor location sensing using active RFID,” Wireless Netw., vol. 10, no. 6, pp. 701–710, 2004. [15] J. Wang, F. Adib, R. Knepper, D. Katabi, and D. Rus, “Rf-compass: Robot object manipulation using RFIDs,” in Proc. ACM MobiCom, 2013, pp. 3–14