Muti-Target Locaization of Breathing Humans ChangKyeong Kim, Joon-Yong Lee, Taechong Cho, Dongbok Ki, Bong Ho Cho, and Jihoon Yoon Dept. of Information & Technoogy, Handong University, Pohang 79-78, Korea Emai: pau3@hgu.edu, joonee@handong.edu, bokmanki@hanmai.net, crissmond@gmai.com Dept. of Nanobio Materias and Eectonics, Gwangju Institute of Science and Technoogy Oryong, Bukgu, Gwangju 5-7, Korea, Emai: taechong@gist.ac.kr Hyundai Mobis, Beon-gi, Mabuk, Giheung, Yongin, 446-9, Korea, Emai: yaudk@naver.com Abstract In this paper, a technique for detecting the ocations of one or more persons, as we as their breath patterns, using an utra-wideband radar sensor network, is proposed. Technica probems that occur in the estimation process, such as fase detection due to indirect refection and ambiguity probem, are introduced. Maximum-ikeihood estimation is suggested as the soution to these probems and tested on sets of radar measurements. I. INTRODUCTION Utra-wideband UWB) signas have been proposed as the most suitabe soution for appications performing accurate distance estimation because of their high resoution. Recenty, considerabe attention has been paid to the area of the detection of bioogica signas, such as those of a breathing pattern or heart rate, by detecting sight movement of a human body. Using the UWB radar, a breathing pattern of one or more persons can be detected aong with distance and/or ocation information, and many studies on this technoogy have been conducted [] [3]. Additionay, there have been studys presenting experimenta resuts on breathing pattern detection in an environment where a wa bocked the space between a person and the UWB radar [4] []. To achieve this objective, the techniques of correation detection, static background remova, and frequency anaysis of breathing patterns have been used. This study aims to detect a breathing pattern of one or more persons who breathe at a fixed position and their ocations in a two-dimensiona space. There are severa technica difficuties in impementing the process. First, it is not easy to remove indirecty refected signas refected from static background aong with a target person this phenomenon is iustrated in section II greater in detai). These signas might be fasey detected as another target. In this paper, a technique for removing these mutipe refection signas invoving the human body that screens them based on distance information and breathing pattern information of the detected potentia targets is proposed. Second, when more than one person s ocations are estimated based on the measured distance information by radar, ambiguity can arise. We used a maximum ikeihood estimation technique to sove this probem. This paper is organized as foows: section II introduces sets of radar measurements and its resuts. A process to detect distance and breathing frequencies of potentia targets from the measured data is aso shown in section II. In Section III, a process for estimating the number and ocations of breathing targets from the detected information is expained. Finay, test resuts of the estimation technique are presented in section IV. II. RADAR MEASUREMENTS UWB radar measurements were carried out using the Puson P4 monostatic radar modue manufactured by Time Domain, Inc. Each radar has two omni-directiona dipoe antennas attached so as to transmit and receive UWB puses. The three radars were arranged in known ocations and fixed on a foam pad to be paced at a height of about.7 m, which is near the average height of the chest a sitting adut. In addition, one to three peope were positioned around the radars so that the breathing pattern and ocation of each person woud be estimated. The ocation of each person was aso determined in advance to evauate the accuracy of the estimation. The reative ocation of each radar and person was measured using a aser distance measuring device. Ten sets of measurements were taken: five sets with one target, four sets with two targets, and one set with three targets. The signa received at the ith radar can be expressed as r i) τ; t), where the superscript i) indicates the index for the radar, τ denotes the propagation deay of a refected waveform and contains the distance information of a target, and t denotes the measurement time. When measuring received signas, each radar adopts an average for the transmission of 496 puses, thereby increasing the signa-to-noise ratio of the received signa. When the person to be detected takes a breath, a portion of body parts such as the chest and/or abdomen aso moves according to a breathing pattern. This can change the structure of a mutipath channe between the transmitting and receiving antennas of the radar, and as a resut, the receiving signa aso changes. Accordingy, by setting a receiving signa at a specific time as a reference signa, the difference between the measured signa and the reference signa is measured so that changes in signa components can be observed by removing static background signas. Let s denote the difference signa at a random moment t by x i) τ; t). For the timing detection of mutipath signa components, matched fitering is performed. If the correator tempate waveform is denoted by s τ), correation function R i) xs,bp Δτ; t) is obtained. Here, subscript BP indicates that the signa has been bandpass fitered. Our anaysis used a Butterworth fiter in the frequency band of.hz to.hz which corresponds 978--4799-968-/3/$6. 3 IEEE 49
a) Fig. : Different kinds of refections. Fig. : a) Experimenta setup forone-person measurement. Pots of b) R 3) S xs,bp r; t) 3). and c) xs,bp r; λ) to a genera human breathing frequency band. In addition, by taking the Fourier transform of R i) xs,bp Δτ; t) with respect to variabe t, spectra density S xs i) Δτ; λ) is obtained. Figure shows the measurement resuts in which one person breathes. Figure -b) shows the pot of R 3), xs,bp r; t) where para r is the vaue that converts Δτ into distance. A periodic signa is observed at 3. m, which is the distance between the radar and the person. Additionay, a simiar movement is detected at 5. m, 6. m, and 6.6 m, which was caused by the b) c) mutipe refection that incudes the target and other objects on its refection path. This phenomenon is aso conspicuousy observed in the frequency domain. Figure -c) shows the pot of S 3). xs,bp r; λ) The breathing frequency and distance of the potentia targets are determined by the foowing procedure. First, vaues of distance r where S i) xs,bp r; λ) dλ has oca peaks in the region where it exceeds the specific threshod are determined. Then, assuming that a vaue of the distance obtained here is denoted by r j, find the vaues of frequency λ where S i) xs,bp r j; λ) has peaks in the region satisfying S i) xs,bp r S i) j; λ) > θ max λ xs,bp r j; λ). The points indicated by the circe in Figure -c) represent the distance and frequency information detected using the method described in the above. Among the severa points detected, ony one contains the information on the actua target distance and breathing frequency, whereas the other points are a erroneousy detected. Figure iustrates this phenomenon. In the figure, path # is the direct path refected by a human body, path # receives static background signas, and path #3 contains indirect refections from a human body and background objects together. When the reference signa is subtracted from the received signa, the indirecty refected signa received via path #3 cannot be removed, whereas the static background signa can be removed. Because of this, a system might fasey detect another target other than the actua target. Figures 3 shows the test environment in which two peope are ocated, and the measurement resuts that were obtained at radar. The two peope breathe at. Hz and.4 Hz, respectivey, and the mutipe signa components at three different ranges are observed. Interestingy, mutipe frequency components are detected at 3.3 m. This is indeed the detection of the harmonic components due to the reguar movements of the human body as a resut of breathing. The vaues detected through the above process at the ith 5
Fig. 3: a) Experimenta setup fortwo-person measurement. Pots of b) R ) S xs,bp r; t) ). and c) xs,bp r; λ) radar can form a matrix R i) = a) b) c) r i), λi) r i), λi).. ) ) ) r i) k i, λ i) k i, ) where each row vector in the matrix indicate a detected point anditisassumedtosatisfyr i) r i) r i) k i, i. The number of observation vectors is denoted by k i. As aready mentioned, matrix R i) can contain signa components received by indirect as we as direct refections from the human body. In order to distinguish these signa components, an initia screening process is performed by anayzing the characteristics of the signa frequency. The rationae is as foows: human body movement due to breathing is significanty sower than the propagation deay of a signa, so it is highy probabe that the movements of the directy and indirecty refected signas from ) the human body are ) synchronized. For exampe, if r i) j, λ i) j and r i), λ i) are vectors detected from the movement of the same person, and r i) j <r i), we can assume that λ i) j λ i). Furthermore, because these two points are where S i) xs,bp r; λ) has oca peaks, it can be assumed that both S i) xs,bp Δτ j; λ j ) and S i) xs,bp Δτ ; λ ) have the same phase or phase difference of as arge as π. Therefore, we can assume that two vectors that satisfy the foowing conditions are due to the movement ) of the same person and thereby remove r i), λ i) : ) λ i) j λ i) <θ λ, ) S i) xs,bp r j; λ j ) ± S i) xs,bp r ; λ ) <θ p, where θ λ and θ p are threshods, respectivey. When more than two observations that satisfy the above conditions exist, a vector detected at the cosest distance is chosen, and the others are removed by an assumption that they were detected because of indirect refections. Now, a new matrix R i) is obtained by the resut, ) r i), λi) ) r i) R i) =, λi)., ). ) r i), k λi) i k i where k i is the number of remaining vectors eft after removing observations due to indirect refections. For exampe, in the experiment shown in Figure, k 3 =4and k 3 =are obtained when θ p = π 5 and θ λ =.4, respectivey. It is noted that the number of the remaining vectors is sti greater than the actua number of targets. The next section presents methods to estimate the number, ocation, and breathing pattern of targets using the matrix R i) obtained from each radar. III. MULTI-TARGET LOCATION If the number of breathing objects that exist around a given radar is n, n satisfies n N = min k i. 3) i 3 If the ocation of the jth target is ϕ j, j n, andthetwodimensiona spatia point in which the ith radar is ocated is α i), the distance between the jth target and ith radar, namey 5
d i) j, can be defined as d i) j = α i) ϕ j. 4) Here, we try } to estimate the ocations of n targets, {ϕ,ϕ,,ϕ n, using observation matrices R i) s. In this process, various compex numbers of cases can occur. Firsty, there is the number of cases in which n vectors are seected from k i observations obtained from the ith radar. If we et this number be Q i,then ) ki Q i =, 5) n and the number of cases that seect n observations from each 3 R i) s becomes Q i. Now we can make n!) different combinations, which is the number of cases that makes n groups comprising three vectors by seecting one vector from each R i). Therefore, when the number of potentia targets is assumed to be n, the possibe tota number of combinations, M n becomes 3 ) ki M n =n!). 6) n If this is expressed using a matrix, it can aso be exhibited as ) ) a ),n,m, b),n,m,, a ) n,n,m, b ) n,n,m ) ) C n,m = a ),n,m, b),n,m,, a ) n,n,m, b ) n,n,m ) ), 7) a 3),n,m, b3),n,m,, a 3) n,n,m, b 3) n,n,m where the ith row is the permutation of n vectors seected from the matrix R i), whie the first row is arranged such that a ),n,m a),n,m... a) n,n,m. Index m is the index that indicates one of M n possibe combinations and satisfies m M n. Therefore, the tota possibe number of the N matrix C n,m becomes M n, which means the number of n= ambiguities. Now, et us estimate the ocation of each target based on matrix C n,m. Define a vector ˆϕ j,n,m as the estimated ocation of the jth target. The estimated range between the ith radar and jth target, ˆdi) j,n,m is given by ˆd i) j,n,m = α i) ˆϕ j,n,m. 8) The error between the estimated and measured distances, can be defined as δ i) j,n,m δ i) j,n,m = ai) i) j,n,m ˆd j,n,m, 9) where, a i) j,n,m is the measured distance between the ith radar and jth target designated by matrix C n,m. When the measurement errors of the distances are assumed to be independent, identicay distributed i.i.d.) random variabes, each of which has f δ δ) as its margina density, the ocation of the jth target can be estimated through maximum ikeihood estimation as foows: ˆϕ j,n,m =argmax ϕ 3 ) f δ a i) α j,n,m i) ϕ. ) If we further assume that f δ δ) = e δ /σ δ, ) σ δ π then east squares estimation can be appied to give ˆϕ j,n,m =argmin ϕ 3 [ a i) α j,n,m i) ϕ ]. ) The breathing frequency of each target can be estimated in a simiar manner. When C n,m is given, if the estimate of the breathing frequency of the jth target is assumed to be ˆλ j,n,m, the error between this estimate and the breathing frequency measured in the ith radar becomes ɛ i) j,n,m = bi) j,n,m ˆλ j,n,m. When the measurement errors of the breathing frequencies are assumed to be i.i.d. Gaussian random variabes with means of and variances of σɛ, ˆλ j,n,m can be obtained by using east squares estimation as foows: ˆλ j,n,m =argmin λ 3 b i) j,n,m λ ). 3) Because the number of combinations is M n, assuming the existence of n targets, the number of estimates that can be obtained of the ocation and breathing frequency of each target becomes M n, respectivey. Among the estimates, one estimate shoud be seected for each para, namey ˆϕ j,n and ˆλ j,n. If M n combinations are assumed to a be equiprobabe, then ˆϕ j,n =ˆϕ j,n,μn and ˆλ j,n = ˆλ j,n,μn,where μ n =argminl n, m). 4) m In the above equation, the cost function L n, m) is defined as ) i) ) i) n 3 δ L n, m) = j,n,m ɛ j,n,m +. 5) σ δ σ ɛ j= Finay, the number of targets, n shoud be determined. Finding the optima vaue of n is a difficut task. When the vaue of n exceeds the number of actua targets, a arge increase in the cost function L n, μ n ) can be predicted. In ight of this, the present study obtained the estimate of n, namey ν, by cacuating the rate of L n, μ n ) according to n and appying a threshod as foows: ν =argmax n [ n L n, μ n) <θ L ], 6) where θ L is the threshod vaue. Hence, the resuting estimates for the ocation and breathing frequency of the jth target can be obtained as ˆϕ j =ˆϕ j,ν and ˆλ j = ˆλ j,ν, respectivey. 5
8 8 6 6 4 4 Radar Radar Target Radar Radar Target Radar 3 Target Radar3 4 4 6 6 4 4 6 a) Estimated ocation the target. 6 8 6 4 4 6 a) Estimated ocations of the targets. ampitude.5.5 5 5 5 3 time sec) b) Breathing pattern of the target. Fig. 4: The estimated ocation and resuting breathing pattern for one-person experiment shown in Figure obtained with σ δ =.5 3, σ ɛ =.5 5,andθ L =. IV. TEST RESULTS As mentioned in the previous section, there exist ambiguities in the process of estimating the ocation and breathing pattern of the targets. By appying the detection technique for the ocation and breathing pattern of the targets presented in Sections III, we can remove these ambiguities. Figure 4 and 5 show the resuts of removing the ambiguity that occurs during the estimation of the ocation and the resuting breathing pattern. Tabe I summarizes the test resuts for the measurement sets. High accuracy of the estimation of the number and ocations of the targets is achieved in spite of a arge number of ambiguities. V. CONCLUSIONS The present study proposed a detection technique for the ocation and breathing pattern of an unknown number of peope. The agorithm proposed in this study was appied to data sets measured in an indoor office environment and exhibited a significanty high eve of estimation accuracy. Among the contents suggested by the study, the process of detecting signa components that occur due to the motions resuting from breathing and, based on its outcome, detecting breathing patterns basicay took a simiar approach to that introduced in existing studies. However, this study has its ampitude ampitude.5.5.5.5 5 5 5 3 time sec) 5 5 5 3 time sec) b) Breathing pattern of target upper) and target ower). Fig. 5: The estimated ocations and resuting breathing patterns for two-person experiment shown in Figure 3 obtained with σ δ =.5 3, σ ɛ =.5 5,andθ L =. unique contribution in terms of using the technique that estimates the number, ocation, and breathing pattern of the targets after removing a number of ambiguities and prevents fase detection caused by factors such as the harmonic components of the breathing frequencies and indirecty refected signas received by successive refections from the human body and the surrounding objects. VI. ACKNOWLEDGEMENTS This research was supported by the Basic Science Research Program through the Nationa Research Foundation of Korea NRF) funded by the Ministry of Education, Science and Technoogy -54). REFERENCES [] M. Baboi, O. Boric-Lubecke, and V. Lubecke, A new agorithm for detection of heart and respiration rate with UWB signas, in 34th Annua Internationa Conference of the IEEE EMBS, pp. 3947 395, Aug.. 53
TABLEI:Testresuts. estimated estimate of breathing measurement number of number of ocation error m) number of frequency Hz) number targets ambiguities targets target target target 3 target target target 3 4.393.3.397. 3.79. 4.39.6 5 4.634.3 6,34.83.34.3.44 7 7,8.87.583.4.7 8 7.35.5..4 9 4.84..4.4 3,,4 3.57.374.334.6.48.8 [] Y. Chen, E. Gunawan, K. S. Low, Y. Kim, C. B. Soh, A. R. Leyman, and L. L. Thi, Non-invasive respiration rate estimation using Utra- Wideband Distributed Cognitive Radar system, in Engineering in Medicine and Bioogy Society, 6. EMBS 6. 8th Annua Internationa Conference of the IEEE, pp. 9 93, Sept. 6. [3] W. Chunming and D. Guoiang, The study of uwb radar ife-detection for searching human subjects, in Internationa Conference on Future Eectrica Power and Energy System, pp. 8 33, Feb.. [4] K. Higashikaturagi, Y. Nakahata, I. Matsunami, and A. Kajiwara, Noninvasive respiration monitoring sensor using UWB-IR, in IEEE Internationa Conference on Utra-Wideband, 8. ICUWB 8., pp. 4, Sept. 8. [5] I.Immoreev, UWB radar for patient monitoring, Aerospace and Eectronic Systems Magazine, IEEE, vo. 3, pp. 8, Nov. 8. [6] O. Kosch, F. Thie, F. S. di Cemente, M. Hein, and F. Seifert, Monitoring of human cardio-pumonary activity by muti-channe UWB-radar, in IEEE-APS Topica Conference on Antennas and Propagation in Wireess Communications APWC), pp. 38 385, Sept.. [7] J. C. Y. Lai, Y. Xu, E. Gunawan, E. C.-P. Chua, A. Maskooki, Y. L. Guan, K.-S. Low, C. B. Soh, and C.-L. Poh, Wireess sensing of human respiratory paras by ow-power Utrawideband impuse radio radar, IEEE Trandsaction on instrumentation and measurement, vo. 6, pp. 98 93, Mar.. [8] A.Lazaro, D.Girbau, and R.Viarino, Anaysis of vita signs monitoring using an ir-uwb radar, Progress In Eectromagnetics Research, vo., pp. 65 84,. [9] M. Leib, W. Menze, B. Scheicher, and H. Schumacher, Vita signs monitoring with a UWB radar based on a correation receiver, in Proceedings of the Fourth European Conference on Antennas and Propagation EuCAP),, pp. 5, Apri. [] K. Ota, Y. Ota, M. Otsu, and A. Kajiwara, Edery-care motion sensor using UWB-IR, in Sensors Appications Symposium SAS),, IEEE, pp. 59 6, Feb.. [] N. V. Rivera, S. Venkatesh, C. Anderson, and R. M. Buehrer, Mutitarget estimation of heart and respiration rates using Utra wideband sensors, in European Signa Processing Conference, Sept. 6. [] J. Sami, S. Sangodoyin, and A. F. Moisch, High resoution para estimation for Utra-Wideband MIMO radar, in Record of the Forty Fourth Asiomar Conference on Signas, Systems and Computers ASILOMAR), Conference, pp. 9 33, Nov.. [3] A. Sharafi, M. Baboi, M. Eshghi, and A. Ahmadian, Respirationrate estimation of a moving target using impuse-based utra wideband radars, Austraas Phys Eng Sci Med, Mar.. [4] M. Y. W. Chia, S. Leong, C. Sim, and K. Chan, Through-wa UWB radar operating within FCC s mask for sensing heart beat and breathing rate, in European Radar Conference, 5. EURAD, pp. 67 7, Oct. 5. [5] B. Levitas and J. Matuzas, Uwb radar for breath detection, in th Internationa Radar Symposium IRS),, pp. 3, June. [6] W. Li, X. Jing, Z. Li, and J. Wang, A new agorithm for through wa human respiration monioring using GPR, in 4th Internationa Conference on Ground Penetrating Radar, pp. 947 95, June. [7] J. Li, L. Liu, Z. Zeng, and F. Liu, Simuation and signa processing of UWB radar for human detection in compex environment, in 4th Internationa Conference on Ground Penetrating Radar GPR),, pp. 3, June. [8] J. Sachs, M. Aftanas, S. Crabbe, M. Drutarovsk, R. Kukas, D. Kocur, T. Nguyen, P. Peyer, J. Rovnakova, and E. Zaikov, Detection and tracking of moving or trapped peope hidden by obstaces using utrawideband pseudo-noise radar, in European Radar Conference, 8. EuRAD 8, pp. 48 4, Oct. 8. [9] S. Singh, Q. Liang, D. Chen, and L. Sheng, Sense through wa human detection using UWB radar, EURASIP Journa on Wireess Communications and Networking, Jan.. [] E. Zaikov, Uwb radar for detection and ocaization of trapped peope, in th Internationa Radar Symposium IRS),, pp. 4, June. 54