Development of an UWB Rescue Radar System - Detecton of Survvors Usng Fuzzy Reasonng - Iwak Akyama Shonan Insttute of Technology Fujsawa 251-8511 Japan akyama@wak.org Masatosh Enokto Shonan Insttute of Technology Fujsawa 251-8511 Japan masa-420@akyama.elec.shonant.ac.jp Akhsa Ohya Graduate School of Systems and Informaton Engneerng, Unversty of Tsukuba Tsukuba 305-8573 Japan ohya@cs.tsukuba.ac.jp Abstract We are engaged n the development of an ultra-wdeband (UWB) rescue radar system for accurately and promptly rescung survvors bured under collapsed houses n cases of dsaster. Snce reflected waves from rubble and those from survvors are mxed n the sgnals receved, t s necessary to separate the two to extract the necessary sgnals. In ths study, fuzzy reasonng s used to classfy the receved sgnals nto the followng three classes: components of reflected waves from survvors, those from statc objects such as rubble, and components manly consttuted of other nose. In order to examne the target-sensng capablty of a radar system ncorporatng ths approach, we carred out experments wth a rubble model targetng subjects. Consequently, the actual poston of the subject matched the poston where the grade for the survvor class peaked, whle the grades for the other classes took small values. Keywords: rubble, tme fluctuaton. 1 Introducton We are engaged n the development of an ultra-wdeband (UWB) rescue radar system for accurately and promptly rescung survvors bured under collapsed houses n cases of dsaster[1]. CW radar systems for searchng human vctms were reported[2]. The radar system now under development s desgned to repeatedly transmt wdeband pulsed waves n frequency ranges from 3 GHz to 6 GHz nto collapsed houses and for searchng for survvors n need of help usng recepton of reflected waves. Snce reflected waves from rubble and those from survvors are mxed n the sgnals receved, t s necessary to separate the two to extract the necessary sgnals. In ths study, fuzzy reasonng s used to classfy the receved sgnals nto the followng three classes: components of reflected waves from survvors n need of help, those from statc objects such as rubble, etc., and components manly consttuted of other nose. The radar system transmts and receves pulses for approxmately 1 mnute, and can store receved sgnals n chronologcal order. Thus, the same delay tmes of the receved sgnals, that s, temporal varatons n reflected components from any object located at a certan dstance from the antenna, are used. Ths study utlzes, as fuzzy parameters, the followng three statstcs values for temporal varatons: mean value, standard devaton and average frequency obtaned from the spectrum. The membershp functons for these three parameters were defned and the grades for each class were deduced. For ths approach, whch we have already proposed for the dfferent applcaton[3], t was decded that the dstrbuton of grades for the three classes n range drectons would be dsplayed wthout makng a fnal classfcaton. Ths s because t was assumed that n a real stuaton n whch a survvor s sought, an overall decson would be made not only based on the radar system results but on other nformaton as well. In order to examne the target-sensng capablty of a radar system ncorporatng ths approach, we carred out experments wth a rubble model targetng subjects. Frst, wth plywood placed about 1 meter n front of an antenna, one subject stood about 1 meter behnd the board. Then, another subject stood further behnd the frst. We carred out experments by
settng the dstance between the two subjects at 1 m n one experment and 3 m n another experment. Then, an antenna was placed about 1 meter above a subject lyng face up, and waves were transmtted down and receved back. Further, we placed concrete blocks between the antenna and the subject, and checked that the waves transmtted through the rubble, reflected off the subject, and could be receved back We appled our approach to the tme-seres sgnals for one mnute that were obtaned from these experments, and calculated the grades for the three classes. Consequently, the actual poston of the subject matched the poston where the grade for the survvor class peaked, whle the grades for the other classes took small values. ths equpment. Both the transmtter and recever were connected to a personal computer through an Ethernet cable, so the personal computer could control wave transmsson and recepton. Table 1 lsts the specfcatons of the EVK system. 2 UWB Rescue Radar System As llustrated n Fgure 1, the UWB (ultrawdeband) radar system used n ths study repeatedly transmts nto and receves UWB pulses from rubble, searchng for a survvor who s present behnd or below t. radar equpment Fgure 1: The Basc prncple of the searchng for survvors usng UWB rescue radar system Personal Computer HUB PulsON 200 EVK transmtter PulsON 200 EVK recever Fgure 2: Block dagram of the expermental UWB rescue radar system Fgure 2 s a block dagram of the system that we prototyped n ths study. As a transmtter/ recever, we adopted the PulsON 200 EVK system of Tme Doman Corporaton n the Unted States[4]. We nstalled a horn antenna to Fgure 3: Horn antenna. Table 1: Specfcatons of PulsON 200 EVK Sze Weght Center frequency Bandwdth Pulse repetton frequency Power consumpton Supply voltage Approx. 19.3 24.1 6.8 cm Approx. 750g Approx. 4.7 GHz Approx 3.2GHz Approx. 9.6MHz Transmsson: 12.2 W Recepton: 11.9 W 7.5 VCD Images were generated by performng envelope detecton on receved sgnals and brghtness modulaton on ampltude, wth dstance beng set n the horzontal drecton and the passage of tme n the vertcal drecton. These mages are called M-mode mages. In such M-mode mages, snce the waves reflected from statc objects do not fluctuate, they are represented as parallel lnes, whle movng objects are seen as non-parallel lne patterns. However, n M-mode mages, f the sgnals receved have strong reflecton components from a statc object, fantly fluctuatng components cannot be observed. Thus, n order to remove any reflecton components due to
statc objects, the Mˆ (t), calculated wth the followng expresson, s dsplayed as an mage [1]. Mˆ (t) r (t) 1 N N r k k= 1 (t) (1) where s the number of pulse recurrences, t s the tme of the receved sgnal, and r (t) s the envelope of the -th (ordnal number) receved pulse wave. 3 Detecton wth the Ad of Fuzzy Reasonng of Reflecton Components from a Survvor In order to detect a survvor n need of help usng a radar system, the authors have nvented an approach usng fuzzy reasonng to detect reflecton components from survvors. Frst, the receved sgnals are dvded nto the followng classes: reflecton components from survvors n need of help (S), reflecton components from statc objects such as rubble (R), and reflecton components nvolvng temporal varatons n whch nose (N) s domnant. The characterstcs of the survvors class are some degree of recepton ntensty, large temporal varatons, and, n partcular, strong perodcty. Those of the rubble class are hgh ntensty and domnant slow fluctuatng components. Those of the nose class are low ntensty and temporal fluctuatons, but no observaton of perodcty. The followng three parameters are used to quantfy the characterstcs of each class: the average value (V m ) of temporal fluctuatons for one mnute at the receved voltage, standard devaton (V SD ), and the average frequency (f m ) obtaned from the spectrum of temporal fluctuatons. Fgure 4 shows the membershp functons for the fuzzy parameters for calculatng the grades for each class: Fgure 4: Membershp functons for the three fuzzy parameters; average value, Standard devaton, and average frequency. From the grades for each class as calculated by the membershp functons shown n Fgure 4, the correspondng mnmum value shall be the grade for that class. 4 Rubble Model Experment We construct a rubble model faclty to represent a collapsed wooden house. In ths rubble model experment, we carred out a lateral search experment and a lower search experment. Fgure 5 s a conceptual dagram of these experments. Lateral search experment Lower search experment Fgure 5 Conceptual dagram of rubble model experment 4.1 Lateral Search Experment In the experment, UWB pulses were transmtted and receved by an antenna facng the front of a pece of plywood. We placed the plywood at a poston about 1 meter from the antenna. In addton to one subject standng about 1 meter behnd the plywood, another subject stood further back at a poston dagonal to the frst subject. The dstance between the two subjects was set to 1 meter n one nstance as shown n Fgures 6 and 3 meters n another nstance as shown n Fgures 6. Fgure 7 shows the measurements for the sgnals receved n the cases of Fgures 6 and.
A dstance of 1 meter between the two persons A dstance of 3 meters between the two persons Fgure 7 Receved Sgnals A dstance of 3 meters between the two persons Fgure 6 Lateral search experment. M-mode mages. Mˆ (t) mages. A dstance of 1 meter between the two persons Fgure 8: M-mode mage and Mˆ (t) mage when the dstance between the two subjects was 1 meter. The dstance s expressed along the horzontal drecton, and the elapsed tme of repeated pulse transmsson s expressed along the vertcal drecton.
Fgures 8 and 9 show M-mode mages and Mˆ (t) mages obtaned through envelope detecton of receved sgnals that were measured for one mnute at 2-second ntervals. In the M- mode mages, reflectons from the plywood were strong and expressed as vertcal parallel lnes. To the left of those lnes, we can see temporal fluctuatons correspondng to the subjects. Although we can observe the temporal fluctuatons of the subject standng n the back n the Mˆ (t) mage of Fgure 8, n Fgure 9 we cannot detect the back subject when standng 3 meters behnd the front subject. meter case and the 3-meter case, respectvely. As shown n the dstrbuton of V SD (Fgure 10 and Fgure 11 ), although we can see a peak at the poston where the subject was standng, there s also a peak at the locaton of the plywood, whch shows that complete separaton from statc objects has not yet been realzed. Arrows n Fgure 10 and Fgure 11 ndcate the poston of the plywood. Then, Fgures 12 and 13 show the determnatons of the grades for the classes as obtaned usng the membershp functons: s the dstrbuton of the grades for survvors, the dstrbuton of the grades for rubble, and the dstrbuton of the grades for nose. The dstrbuton of the mnmum values of the grades obtaned from the dstrbutons of, and are shown by (d) and thus represent the fnal grades for the three classes. M-mode mages Fgure 10: Dstrbuton of Fuzzy Parameters: In the 1-meter Case. s V m,, V SD, and f m. Mˆ (t) mages Fgure 9 M-mode mage and Mˆ (t) mage when the dstance between the two subjects was 3 meters. The dstance s expressed along the horzontal drecton, and the elapsed tme of repeated pulse transmsson s expressed along the vertcal drecton. Fgures 10 and 11 show the dstrbutons of the fuzzy parameters V m, V SD, and f m n the 1- Fgure 11: Dstrbuton of fuzzy Parameters: In the 3-meter Case. s V m,, V SD, and f m.
Fgure 12: Dstrbuton of grades for each class as determned by the membershp functons n the 1-meter case. s for survvors, for rubble, and for nose. V m s expressed as sold lne, V SD as dotted lne, and f m as dash-dot lne. (d) s the dstrbuton of the mnmum values obtaned from the each dstrbutons of grades. The dstrbuton of grades for survvors s expressed as dash-dot lne, for rubbles as dotted lne, and for nose as sold lne. (d) (d) poston as shown by arrows. In contrast, the grades for the other classes are small. Ths shows that t s most lkely that one survvor each exsts at the 2-meter poston and the 3- meter poston, whch matches the postons where the subjects were actually standng. As shown n Fgure 13 (d), we notce that n the survvor grade, there are peaks at the 3-meter poston, 4-meter poston and 5-meter poston, n addton to the 2-meter poston. However, at the 3-meter and 4-meter postons, the nose grade s also hgh, whch shows that t s most lkely that survvors are present at the 2-meter and 5-meter postons as shown by arrows, matchng the postons where subjects were actually standng. 4.2 Lower Search Experment We conducted a lower search experment assumng the case of a person bured under a collapsed house. We nstalled an antenna unt downward, wth a subject lyng underneath. Fgure 14 s a photograph showng how the experment was conducted. Fgure 15 shows M-mode mages and Mˆ ( t) mages. There were strong reflectons from the block and we were unable to observe reflectons from the subject who was beneath t. Fgure 16 shows the dstrbutons of the fuzzy parameters. Lookng at the dstrbuton for V SD, we cannot fnd a clear peak attrbutable to the subject because reflectons from the block are so strong. Fgure 13: Dstrbuton of grades for each class as determned by the membershp functons n the 3-meter case. s for survvors, for rubble, for nose. V m s expressed as sold lne, V SD as dotted lne, and f m as dash-dot lne. (d) s the dstrbuton of the mnmum values obtaned from the each dstrbutons of grades. The dstrbuton of grades for survvors s expressed as dash-dot lne, for rubbles as dotted lne, and for nose as sold lne As shown n Fgure 12 (d), we notce that n the grade for survvors n need help, peaks can be seen at the 2-meter poston and the 3-meter Fgure 14: Lower Search Experment Fgures 17, and show the grades for each class, and Fgure 17 (d) s the plot of the mnmum values of the grades. Fgure 17 (d) shows that n the survvor grade, there s a peak
at the 1.5-meter poston as shown by an arrow. Ths poston matches the actual poston of the subject. (d) Fgure 17: Dstrbuton of grades for each class as determned by the membershp functons. s for survvors, for rubble, for nose. V m s expressed as sold lne, V SD as dotted lne, and f m as dash-dot lne. (d) s the dstrbuton of the mnmum values obtaned from the each dstrbutons of grades. The dstrbuton of grades for survvors s expressed as dash-dot lne, for rubbles as dotted lne, and for nose as sold lne. Fgure 15: M-Mode Image and ˆ ( t) mage. M Fgure 16: Dstrbuton of fuzzy parameters. s V m,, V SD, and f m. Fgures 18 to 20 show the results for the case n whch the subject took a deep breath n the experment shown n Fgure 14. Fgure 18 shows the M-mode mage and Mˆ ( t) mage. Fgure 18: M-Mode Image and Mˆ ( t) Image.
Fgure 19: Dstrbuton of fuzzy parameters. s V m,, V SD, and f m. (d) 5 Concluson Wth the ad of receved sgnals usng the fuzzy reasonng of a UWB rescue radar system, we establshed an approach to separate reflecton components of rubble and other obstructons and extracted the reflecton components of persons posng as survvors. Usng ths approach, we conducted rubble model experments and examned the target-sensng capablty of the radar system. As a result, we were able to successfully dstngush a subject standng 1 meter behnd plywood and another subject standng 3 meters as well as 1 meter behnd the frst subject. When we appled ths approach n condtons n whch a subject lyng on hs back was breathng quetly as well as breathng deeply, we succeeded to detect the poston of the subject. Acknowledgements We conducted part of ths study wth assstance from the Specal Project for Earthquake Dsaster Mtgaton n Urban Areas of the Mnstry of Educaton, Culture, Sports, Scence and Technology. References Fgure 20: Dstrbutons of Grades for Each Class. s for survvors, for rubble, for nose. V m s expressed as sold lne, V SD as dotted lne, and f m as dash-dot lne. (d) s the dstrbuton of the mnmum values obtaned from the each dstrbutons of grades. The dstrbuton of grades for survvors s expressed as dash-dot lne, for rubbles as dotted lne, and for nose as sold lne. Fgure 19 shows the dstrbutons of the fuzzy parameters, and Fgure 20 shows the dstrbutons of the grades for each class. In the Mˆ ( t) mage of Fgure 18, we can see temporal fluctuatons at the poston where the subject was lyng. Among the fuzzy parameters shown n Fgure 19, the V SD value s hgh. As shown n Fgure 20, we see a peak grade value for the survvor class at about the 1-meter poston as shown by an arrow. Ths poston matches the actual poston of the subject. [1] I. Akyama, Y. Arak, M. Isozak, M. Ohk, A. Ohya (2004): UWB Radar System Sensng of Human Beng Bured n Rubbles for Earthquake Dsaster, Abstract of EUROEM2004, 163-164, July 2004. [2] K-M.Chen, Y.Huang, J. Zhang and A.Norman: Mcrowave lfe-detecton systems for searchng human subjects under earthquake rubble or behnd barrer, IEEE Trans. on Bomedcal Engneerng, Vol.27, No.1, 105-114, 2000. [3] I.Akyama, Y.Kawahara, G.Ohash, K.Omoto, K.Itoh, X.Cheng, and A.Ohya (2000) Dagnostc system for screenng of breast cancer usng the most traned neural network and partally traned neural network, Proceedngs of the conference IPMU'2000, volume 1, pages 13-18, Madrd, Span July 2000. [4] http://www.tmedoman.com/fles/products/ P200EVK.pdf