Decision-Making Level Fusion Based on DSmT for Multi-Sensor Life Detection Platform

Transcription

1 enor & Tranducer 204 by IFA Publihing,. L. Deciion-Making Level Fuion Baed on DmT for Multi-enor Life Detection Platform Jingong YANG, Xun CHEN Department of Diater Prevention Intrument, Intitute of Diater Prevention, 60520, China Received: 7 November 203 /Accepted: 28 January 204 /Publihed: 28 February 204 Abtract: Life detection baed on a ingle type of information ource detection technology cannot completely meet the need of the earthquake relief. Firt, we analyze the preent ituation of typical life detection technology, then the advantage of multi-enor ynergy detection technique are given. econd, we explained the platform tructure and information fuion model. Finally, the fuion rule baed on Dezert-marandache theory (DmT) i ued to achieve the deciion level fuion. We preent the reult reached by the fuion with the conjunctive rule, the Dempter, the Yager and the DmT. The experiment how that DmT can get more comprehenive and more accurate information about the cene of recue and urvivor. The multi-enor ynergy life detection platform baed on DmT i more uitable for complex quake recue environment. Copyright 204 IFA Publihing,. L. Keyword: Life detection, Deciion level fuion, Dezert-marandache theory, Infrared enor, Acoutic wave enor.. Introduction earching for urvivor after a diater i alway a race againt time. The correct ue of recue method and equipment to be trapped recue peronnel poitioning, can be twice the reult with half the effort. Exiting life detection technique are baed on a ingle type of enor, there exit their defect. Baed on multi-enor ynergy reearch will become the focu of detecting technology. At preent, the application of multi-enor ynergy mainly concentrated in military, aeropace and indutrial production area, while to detect the recue of life i very little [, 2]. Data and information fuion clearly i a key enabler in the proviion of deciion quality information to the deciion maker. The eence of deciion-making in civilian, military and public ecurity operation i people making timely deciion in the face of uncertainty, and acting on them. Thi proce ha been immeaurably complicated by the overwhelming and increaing volume of raw data and information available in the current age. Multi enor ynergy life detection i to make urvivor recognition baed on multi-ource information fuion. It i more effective than ingle ource detection (uch a audio life detection, video life detection, radar Life Detection, and o on). D- theory i uitable for without any prior information, but alo in the repreentation of uncertainty, meaurement and ha the advantage of combination, and it accord with the deciion the proce of human reaoning. Evidence theory but in evidence for the occurrence of high conflict ituation will be effective fuion have counter intuitive concluion port J. to realize multi-ource information in highly conflict evidence i an urgent problem to be olved. A lot of literature that thi i caued by the Article number P_829 3

2 combination of rule and improved, but the effect i not very ideal. In thi cae, Dezert and marandache et al propoed DmT in DmT i the extenion of the claical D- theory, but DmT i different with the D- theory [3-5]. Thi paper decribe DmT and it application in urvivor identification, and verified with an example. 2. Life Detection Technology Analyi Life detection i to ue the enor technology to collect the urvivor phyiological, phyic and chemical information, to effectively identify the poition of the living organim [6]. According to the enor type can be divided into audio life detection technology (including ound wave, vibration wave), video life detection technology (including optic, optical fiber, infrared) and radar life detection technology (including imaging, the imaging). The life detection technologie baed on ingle type of enor have their own defect. 2.. Audio Life Detection Technology Audio life detection technology i the application of the principle of vibration wave ound pread, through the detection of underground weak uch a the men groan, hout, creeping, beating from audio ound wave and vibration wave to judge whether life exit [7]. But it i a kind of paive detector receiving audio ignal and the vibration ignal; therefore, it i vulnerable to the urrounding noie interference. People need to find pace tretch into probe in the ruin, detection at a lower peed Video Life Detection Technology Video life detection technology i turning recue environment and object into viual image, o a to provide the recuer with environmental information, the urvival poition information, etc. Among them, the infrared detection equipment life in recue in the proce of uing more widely, it will be a different temperature object inviible infrared ray into a viual image, uing the human body black and white image to determine the urvivor and victim of the intrument can be dark, moke to earch the environment, and ha the function of night viion, thi make life work from the light detection effect, can help recue team quickly, accurately determine buried in the ruin in or hidden in the back of the victim of the haze poition [8]. But the infrared life detector could not penetrate obtacle, till need to looking for the big or mall hole, crack, etc, to reach into the probe to oberve internal ituation, o the application ubject to certain retriction Radar Life Detection Technology Radar detection technology i to fue radar technology and biomedical engineering technology to one life detection equipment, with trong penetration, precie range and trong anti-interference ability, trong ability to target recognition etc, which can accurately identify the living organim [9]. Commonly ued radar life detection technology can be claified into two kind, namely Continuou Wave (CW) radar detection life and the Ultra Wide Band (UWB) radar life detection. Continuou wave radar detection mean life continuou wave radar launch mw beam of the body, the reflection of the echo wa living life activitie (uch a breathing, heartbeat, etc) caue budge the modulation, make ome of the echo ignal parameter change, ue of appropriate ignal proceing technology of the parameter of the living organim extract ignal (uch a breathing, heart rate, etc). Ultra-wide band radar detection mean life pule declined beam irradiation living organim, organim the reflection of the echo pule repeated equence of cycle port becaue organim and change information, uing a digital ignal proceing technology of the parameter of the living organim extract ignal (uch a breathing, heart rate, etc). The diadvantage of radar life detector are high cot and electromagnetic wave radiation effect to human body, o radar life detector doe not apply to cloe detection The Advantage of Heterogeneou enor ynergy Viible, the life detection technologie baed on a ingle type of information ource have their own defect, cannot completely meet the need of the earthquake relief, the life detection technology baed on heterogeneou enor become a new hotpot. The target compound recognition baed on information fuion uing heterogeneou enor to collaborative detection ha the following advantage: ) Broaden the urveillance pace and detection coverage; 2) Play to the advantage of each enor and complement each other in order to improve the target recognition rate; 3) The anti-interference performance of the multienor much better than ingle enor; 4) Improve the reliability and fault tolerance. For more collaborative detection enor at home and abroad reearch mainly focu on three apect: The contruction of the multi-enor integration model; Multi-enor election trategy and Control tructure reearch [0] and the multienor information fuion method reearch []. 32

3 3. Multi-enor Life Detection Platform Deign Infrared and acoutic enor can be in the dark, meary wait for the bad environment under the working condition of, and ha imple tructure, low cot, adaptable, doe not damage eaily, etc. Through the infrared enor can get living organim of the environment which image, uing the principal, paive ound wave to living organim can enor for a weak ditre ignal detection of living organim and more accurate poitioning, and both can work together to make the recue plan to provide more accurate bai for deciion-making, reduce relief time, reduce lo diater. 3.. The Platform tructure Thi ynergy detection platform function model i hown a Fig.. 2) Video acquiition module; 3) ignal fuion proceor (FPGA + DP); 4) Embedded proceor: communication, control, diplay, etc. The hardware configuration of infrared and acoutic life ynergy detection platform i hown a Table. Table. The hardware configuration. No. Decription Technical pecification Multienor fuion ignal proceing platform Vibration enor Audio enor ) Backplane; 2) 8 channel ound/vibration ignal ynchronou ampling rate: k/; 3) Embedded proceor: 2.53 GHz Intel Core 2 Duo T9400; 4) ytem throughput GB/, ingle lot throughput: 250 MB/; 5) channel Camera Link interface Impact reitance value >000 g Frequency repone: ~3000 Hz Frequency repone: 200 Hz to 3000 Hz Multi-channel data collection realize low frequency ound and ignal and ultraonic ound wave ignal ynchronou collection. Video collection module implement video collection of the infrared tranmiion and pretreatment. ignal fuion proceor can complete the mater of paive acoutic ignal proceing. Embedded proceor i reponible for the communication of the whole ytem, control, diplay, etc The Platform Building Fig.. Function model of life detection ynergy platform. Firt, the paive acoutic detection technology, in the ruin in the ruin or empty detection, acquiition and extracting the organim of a ound wave ignal (the men groan, hout, creeping, beating from audio ound wave), to judge the living organim exit upected area. In thi area for the ruin and empty, uing infrared viion detecting, living environment image collection. At the ame time to target launch ultraonic active organim, in order to accurately poition are living organim. Through information fuion proceing of the above can realize the target recognition and living location, and provide deciion upport for formulate effective recue plan Hardware tructure The hardware of the platform mainly include: ) Multi-channel ound wave information collector; The ynergy detection platform i compoed of hardware platform and oftware platform. The hardware platform include infrared enor, vibration enor, ultraonic enor, multi-channel acoutic/vibration data acquiition card, ultraonic data collection card, embedded proceor and I/O device, a hown in Fig. 2. The oftware platform ue the LabView a a development tool, to complete the video and ound information collection, information diplay and proceing. embedded proceor Fig. 2. Hardware platform tructure. all kind of acquiition card 33

4 4. Deciion-making Level Fuion Baed on DmT The infrared image and acoutic information i completely different qualitative information, that i, different information. Uually, different information ha the very trong complementary to each other, o the fuion of information produced by the more practical value. However, alien enor information fuion are faced with many difficultie, the main difficulty i currently there i no general mathematical tool and method, decription and analyi of the different information feature and many ource of information fuion correponding. The multiple ource information fuion general can carry out in different level, uch a data level, characteritic level and deciion level. Becaue different inherent characteritic of information and i currently in the data layer of fuion extremely hard, general only limited characteritic layer or deciionmaking for reearch. And deciion-making level fuion i in the top of the aid information on the fuion proceing. 4.. Background on Combination Rule The hybrid Dm rule can deal with the potential dynamicity of the frame and it model a well [5]. The DmT deal properly with the granularity of information and intrinic vague/fuzzy nature of element of the frame to manipulate. The baic idea of DmT i to define belief aignment on hyper-power et D (i.e. free Dedekind lattice) and to integrate all integrity contraint (excluivity and/or non-exitential contraint) of the model, ay M(), fitting with the problem into the rule of combination. Mathematically, the hybrid Dm rule of combination of 2 independent ource of evidence i defined a Eq. for all X D. X X X 2 X X m M 3, () where Φ(X) i the characteritic non-emptine function of a et X, i.e. Φ(X)= if X Ø; and Φ(X) = 0 otherwie, where Ø { Ø M, Ø }. Ø M i the et of all element of D which have been forced to be empty through the contraint of the model M and Ø i the claical/univeral empty et. (X), 2 (X) and 3 (X) are defined by Eq. 2, Eq. 3 and Eq X m X X, X 2,, X D i X X X X 2 X m i X i X, X 2,, X u X u i X I, t i i, (2) (3) 3 X X X X D i, 2,, X X 2 X A X X 2 X m i X i, (4) with u u(x)u(x2) u(x) where u(x) i the union of all θi that compoe X and It θθ2 θn i the total ignorance. (A) correpond to the claic Dm rule for k independent ource baed on the free Dm model Mf (); 2(A) repreent the ma of all relatively and abolutely empty et which i tranferred to the total or relative ignorance aociated with non exitential contraint (if any, like in ome dynamic problem); 3(A) tranfer the um of relatively empty et directly onto the (canonical) dijunctive form of non-empty et Fuion Example of Infrared Image and Acoutic Information In thi cae, the pace of dicernment repreent the urvivor identification on infrared image and acoutic information. The expert give their perception and belief according to their certainty. Conequently we have to take into account all thee apect of the application. In order to implify, we conider only three clae in the following: urvivor a A, no-urvivor, referred a B, and indetermination, referred a C. Hence, on certain tile, A, B, and C can be preent for one or more expert. The belief function have to take into account the certainty given by the expert (number in [0, ]) a well a the proportion of the kind of urvivor identification. We give the obtained reult on a multi-enor life detection ytem for the fuion of two expert in infrared image and acoutic information. Table 2 how the generalized confidence given by two expert, which m : confidence given by expert baed on infrared image; m 2 : confidence given by expert baed on acoutic information; m 3 : confidence given by expert2 baed on infrared image; m 4 : confidence given by expert2 baed on acoutic information. Table 2. Generalized confidence function. Target type m m2 m3 m4 A B C We preent here the reult reached by the fuion with the conjunctive rule, the Dempter, the Yager and the DmT, hown a Table 3. The reult in Table 3 how that Dernpter reult i contrary to intuition, which determine the target identity a "B". Becaue evidence 2 ha "veto", no amount of evidence to upport "A", it i a eriou 34

5 error. In the ame way, Yager rule aign all conflict to "C", which i no good to make a deciion. However, DmT determine the target identity a "A", which i atified the verdict. It mean that DmT i uit to olve the problem brought by high conflict evidence. Table 3. Reult with different conjunctive rule. Reult Dempter Yager DmT m(a) m(b) m(c) m(ab) m(ac) m(bc) 0.67 m(abc) m() Concluion At preent life detector are till baed on a ingle type enor. While in the actual earthquake relief work, life detector and recue often need to combine variou information. Infrared and acoutic ynergy life detection baed on DmT can make up the defect of ingle enor life detection. Deciionmaking level fuion baed on DmT i more uitable for complex environment of earthquake relief, and can get more comprehenive, more accurate information of urvivor. We have hown on real data that there i a difference of deciion following the choice of the combination rule. The reult how that DmT i uit to olve the problem brought by high conflict evidence. Acknowledgement Thi work wa funded by the teacher' cientific reearch fund of china earthquake adminitration (Grant No. 2002); pecial Fund of Fundamental cientific Reearch Buine Expene for Higher chool of Central Government (Project for creation team) (Grant No. ZY20004). Thee upport are gratefully acknowledged. Reference []. Q. J. Zhao,. F. Chen, Multi-ource information fuion method baed on texture analyi, Information and Control, Vol. 36, Iue 2, 2007, pp [2].. Makell, A Bayeian approach to fuing uncertain, imprecie and conflicting information, Information Fuion, Vol. 9, Iue 2, 2008, pp [3]. L. A. Zedeh, Review of hafer, A mathematical theory of evidence, AI Magazine, Vol. 5, Iue 3, 984, pp [4]. R. Yager, On the Dempter hafer framework and new combination rule, Information cience, Vol. 4, Iue 2, 987, pp [5]. F. marandache, J. Dezert, Advance and application of DmT for information fuion, American Reearch Pre, Rehoboth, [6]. K. M. Chen and D. K. Mira, An X-band microwave life detection ytem, in Proceeding of the 6 th Annual Meeting of Bioelectromagnetic ociety, Atlanta, GA, 5-9 June 984. [7]. C. B. Wang, Y. Guo, J. Wang, Analyi of victim location ytem ignal in earthquake diater baed on acoutic and eimic wave detection, Chinee Journal of Engineering Geophyic, Vol. 2, Iue 2, 2005, pp [8]. J. hen, Miniature ound wave life detection ytem for pot-diater relief, Chengdu Univerity of Technology, Chengdu Intitute of Technology, Chengdu, [9]. J. Han, C. Nguyen, Development of a tunable multiband UWB radar enor and it application to uburface ening, IEEE enor Journal, Vol., 2007, pp [0]. B. Lohman, O. Boric-Lubecke, V. M. Lubecke, et a., A digital ignal proceor for Doppler ening of vital ign, torming Media, Wahington DC, 200. []. K. P. Polly, K. D. Chritopher, Tet bed robot development for cooperation clearance, International Journal of Robotic Reearch, Vol. 8, 999, pp Copyright, International Frequency enor Aociation (IFA) Publihing,. L. All right reerved. ( 35