Your SlideShare is downloading. ×
Y.-W. Bai et al.: Design and Implementation of a Home Embedded Surveillance System with Ultra-Low Alert Power 153Manuscrip...
154 IEEE Transactions on Consumer Electronics, Vol. 57, No. 1, February 2011Fig. 2. The home embedded surveillance system ...
Y.-W. Bai et al.: Design and Implementation of a Home Embedded Surveillance System with Ultra-Low Alert Power 155sensing a...
156 IEEE Transactions on Consumer Electronics, Vol. 57, No. 1, February 20110 20 40 60 80 1000102030405060708090100Single ...
Y.-W. Bai et al.: Design and Implementation of a Home Embedded Surveillance System with Ultra-Low Alert Power 1579 shows t...
158 IEEE Transactions on Consumer Electronics, Vol. 57, No. 1, February 2011the same time the MCU maintains a speed faster...
Y.-W. Bai et al.: Design and Implementation of a Home Embedded Surveillance System with Ultra-Low Alert Power 159[12] Yu-K...
Upcoming SlideShare
Loading in...5

Design and implementation of a home embedded surveillance system with ultra low alert power


Published on

For more projects visit @

Published in: Technology, Business
  • Be the first to comment

  • Be the first to like this

No Downloads
Total Views
On Slideshare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Transcript of "Design and implementation of a home embedded surveillance system with ultra low alert power"

  1. 1. Y.-W. Bai et al.: Design and Implementation of a Home Embedded Surveillance System with Ultra-Low Alert Power 153Manuscript received 12/26/10Current version published 02/21/11Electronic version published 02/21/11. 0098 3063/11/$20.00 © 2011 IEEEDesign and Implementation of a Home EmbeddedSurveillance System with Ultra-Low Alert PowerYing-Wen Bai, Zi-Li Xie and Zong-Han LiAbstract — In this paper we design and implement a homeembedded surveillance system with ultra-low alert power.Traditional surveillance systems suffer from an unnecessarywaste of power and the shortcomings of memory conditions inthe absence of invasion. In this design we use PyroelectricInfrared sensors (PIR) and pressure sensors as the alertgroup in windows and doors where an intruder must passthrough. These low-power alert sensors wake up the MCU(Micro Controller Unit) which has power management for theultrasonic sensors and PIR sensors indoors. This statetransition method saves a large number of sensors requiredfor the alert power. We also use the Majority VotingMechanism (MVM) to manage the sensor groups to enhancethe probability of multiple sensors sensing. After the MCUsends the sensor signals to the embedded system, the programstarts the Web camera. Our sensing experiment shows that wereduce the system’s power consumption1.Index Terms — Embedded Surveillance System, MajorityVoting Mechanism, PIR Sensor, Ultrasonic Sensor, Low-PowerState.I. INTRODUCTIONThe traditional surveillance systems take a long time todetect whether there is any intruder. If there is no intruder,the sensing device which continuous to work and consumesmuch power [1-5]. To meet the increased requirements ofthe IEA we have to reduce the standby power of eachelectrical apparatus to less than 1 Watt [6-8]. A recentlypublished survey shows that various attempts have beenmade to reduce such power loss by to making the adaptersmore efficient [9-12]. Another way to improve powerefficiency is accurate control of the apparatus by bothsoftware and microcontroller [13-15].In this paper the alerting sensors with low-powerconsumption are placed near those home windows and doorswhere an intruder must pass through. When an intruderenters the sensing area, the sensors wake up the sleepingMCU (Micro Controller Unit) which starts the power supplyfor the indoor sensors and for the sensor signal transmissionto the embedded system. For the indoor sensors we use the1Ying-Wen Bai is with the Department of Electrical Engineering andGraduate Institute of Applied Science and Engineering, Fu-Jen CatholicUniversity, Taipei, Taiwan, 242, R.O.C. (e-mail: Xie is a graduate student of Fu-Jen Catholic University, Taipei,Taiwan, 242, R.O.C. (e-mail: Li is a graduate student of Fu-Jen Catholic University, Taipei,Taiwan, 242, R.O.C. (e-mail: to improve the sensing reliability [16-17]. Theembedded surveillance system determines the sensor resultsand then decides whether to start the Web camera to bothcapture images and upload these captured images to the Webpage through the Internet [18-20]. We use the MCU’s sleepmode to reduce the alert power consumption for our homeembedded surveillance system when there is no intruder soas to improve the traditional surveillance system withoutwasting the power.II. SYSTEM ARCHITECTUREFig. 1 shows the home embedded surveillance systemwhich has two groups of sensors, indoor and outdoor. Theoutdoor sensor group contains a number of PIR and pressuresensors placed near windows and doors of a home. When theoutdoor sensors sense an intruder, the MCU is woken up andturns on the power for the indoor PIR and ultrasonic sensorsfor the Majority Voting Mechanism. When this is completed,the decision signal passes to the embedded board GPIO(General purpose input and output). The software module ofthe embedded board turns on the Web camera to captureimages, and the embedded system uploads them by means ofthe Web server through the Internet. The user can view theimages captured by the home surveillance system through theInternet. Fig. 2 shows the system architecture.Fig. 1. The state transition for the home embedded surveillance system tosave power.
  2. 2. 154 IEEE Transactions on Consumer Electronics, Vol. 57, No. 1, February 2011Fig. 2. The home embedded surveillance system with ultra-low alertpower.A. Software ModulesOur design shown in Fig. 3 consists of the internal MCUsoftware module and the home embedded system softwaremodule. In the MCU software module, when an intruder hasbeen detected, the MCU wakes up and decide whether to testthe threshold, and then turn on the power supply for theindoor sensors. If the indoor sensors detect no intruder whenthe outdoor sensors are misjudging, the MCU turns off thepower for the indoor sensors and goes back to the alert state.Fig. 3. The software flowchart of the MCU.We use Embedded Linux in the embedded board software.The software flowchart shown in Fig. 4 scans the embeddedboard GPIO pin, and when any input pin is high, thisrepresents a sensor sensing an intruder. The embedded systemturns on the Web camera to capture images.Fig. 4. The software flowchart of the embedded board.B. Hardware ModulesTo reduce the power consumption of the alert state weutilize both pressure sensors and PIR sensors. We use the PIRsensors to detect the change in environment temperaturethrough human temperature in our experiment. The sensingdistance of a PIR sensor can reach 4m, and the sensing angleis 110°. We use a common PIR sensing IC. The standbypower is only 1.05mW if there is no intruder. Fig. 5 shows asystem block diagram of the PIR module.Fig. 5. System block diagram of the PIR module.We place the PIR sensor groups on the ceiling, and thesensing area is a cone-shaped projection area. Covering the
  3. 3. Y.-W. Bai et al.: Design and Implementation of a Home Embedded Surveillance System with Ultra-Low Alert Power 155sensing area with multiple PIR sensor groups extends thesensing area, and by using MVM we enhance the overallsensing probability. But the PIR sensor cannot sense a low-speed or heat-insulated object, because obviously thesewon’t change the environment temperature. The pressuresensors are designed by linear potentiometers. The pressuresensors used are thin and placed on the ground. When anintruder invades, the pressure on the linear potentiometerincreases the resistance value, and the MCU determinesaccording to the resistance value whether someone ispassing through. Fig. 6 shows the system block diagram ofthe pressure sensor module.Fig. 6. System block diagram of the pressure sensor module.For the detection state we use PIR sensors and multi-frequency ultrasonic sensors. Fig. 7 shows the ultrasonicsensor module which uses a typical oscillator chip to designa square waveform generator and to adjust the resistance andcapacitance so as to generate a multi-frequency ultrasonictransmission. The ultrasonic transducer transforms thevoltage waveform into an ultrasonic transmission and thetransducer of the receiver transforms the ultrasonictransmission into the voltage waveform. Since the receivermay experience external interference at different frequencies,it is necessary to screen both the filter signals outside thereceiving frequency and the signal input to the amplifier andthe comparator; as other ultrasonic sensors are alsosusceptible to refractive interference, so we use severalultrasonic sensors at the receiving end. The total number ofultrasonic sensors as the majority of the sensors triggered isthen sent to the MCU.Fig. 7. A system block diagram of the ultrasonic sensor module.Table I shows the sensing characteristics of the PIR sensor,the ultrasonic sensor and the pressure sensor.TABLE ICOMPARISON OF THE CHARACTERISTICS OF PIR SENSOR, ULTRASONICSENSOR AND PRESSURE SENSORSensorConditionfor triggerInfluence ofenvironmenttemperatureSensingtypeDisadvantagesPIRsensorTemperaturechangedDependenceProjectionareaSlow speed andheat insulationUltrasonicsensorMoving block IndependenceLinedirectionNoiseinterferencePressuresensorPressurechangeIndependenceLinedirectionPlacementIII. MAJORITY VOTING MECHANISMWe first discuss the definition and the probability equationof a majority voting mechanism. According to our MVM theresolution count must be greater than n5.0 , n being the totalnumber of sensors. To fit the extreme value of n we use nwto deduce the relation between mmultiple PnP )( and ssingle PP  inthe extreme value of n [8].nwrrssnwsnwsmPPnwrnPPP)1(0)1(])1([)1()( (1)We define}])1[(,]1)1[(,,2,1,0{,)1()( nwnwkandPPnwknkf kss As we expect that nwkkf)1(0)( will converge, we need todetermine whether )( kf is a decreasing function. From theratio test for the convergence function we learn that increasingthe n value gradually decreases the ratio for each )( kf . Therelationship is as follows:]1)1([])1([]2)1([]1)1([)2()3()1()2()0()1(1nwfnwfnwfnwfffffffLet n , so1)1()1()0()1( ssn PPwwfflimWe let 12/1  w and wPs 1)()()()1()()1(0)1( wPPkfPPPsmnnwrnwsnwsm lim (2)If we let mP represent the miss rate of sensors, we can learnthat mm PP 1 , and mP =0 in this case.We rewrite (1) by letting wPs  and thus deduce mm PP 1by the ratio test.1)()()()1()()1(0)1( wPPkgPPPsmnnwrnwsnwsm lim (3)According to (2) and (3) the sensing probability of multiplesensors must be greater than the sensing probability of anysingle sensor. We know that when singleP is greater than 0.5, the)(nPmultiplewill be greater than 0.5. Fig. 8 shows theimprovement of the sensing probability of multiple sensorsthrough majority voting. We obtain Table II by given singleP .If the sensing probability of a single sensor is 0.7, the sensingprobability of seven sensors will be 0.874.
  4. 4. 156 IEEE Transactions on Consumer Electronics, Vol. 57, No. 1, February 20110 20 40 60 80 1000102030405060708090100Single Individual Module Recognition Probability ( % )CombinedSystemOverallRecognitionProbability(%)n = 1n = 3n = 5n = 7n = 9n = 39n = 99n = 159n near to infinityCombined Probability in AnyOdd Number Modules WillAlways be Equal to 50%-Multiple(7)=87.4%-Multiple(5)=83.7%-Multiple(1)=70.0%-Multiple(9)=90.1%-Multiple(3)=78.7%-Multiple(1)=30.0%-Multiple(3)=21.6%-Multiple(5)=16.3%-Multiple(9)=9.88%-Multiple(7)=12.6%Fig. 8. Sensing probability of a single sensor and multiple sensors.TABLE IIRELATIONSHIP BETWEEN singleP AND )(nPmultipleIV. MAJORITY VOTING MECHANISM ANDCOMPLEMENTSIn this Section we continue to deduce the majority votingmechanism and look at the complementing of multiple groupsof sensors. We choose three sensors in a sensor group becausethree is the smallest number possible for the majority votingmechanism, and we use three as an example to show theprocess. If we assume that the sensing probabilities of eachsingle sensor are PPsingle  and the miss rates are P1 , theprobability of three sensors which are all in favor is3PPPP  , and the probability of two sensors in favor andone sensor not in favor is )1(33)(1 2PPPPP  . The entireprobability of a majority vote in favor is the sum of two of theitems mentioned above, which is)1(3 23sensors3 PPPPmultiple (4)We deduce further the sensing probability of a two-set or athree-set sensor group. As mentioned above, the sensingprobability of a single group is (4) and that of three groupscovering the sensing areas is (5).     )1(31)1(33)1(3)1(32322332323sensorsgroup3PPPPPPPPPPPPPmultiplemultiplemultiplegroup(5)Table III shows the overall sensing probability with respectto the different counts of the sensors in a sensor group.TABLE IIIOVERALL SENSING PROBABILITIES AND COUNTSsingleP)(nPmultiplen=3 n=5Single Three Five Single Three Five0.30 0.2160 0.1198 0.0709 0.1631 0.0711 0.03350.40 0.3520 0.2845 0.2382 0.3174 0.2383 0.18680.50 0.5000 0.5000 0.5000 0.5000 0.5000 0.50000.60 0.6480 0.7155 0.7617 0.6826 0.7617 0.81320.70 0.7840 0.8802 0.9290 0.8369 0.9289 0.9665We add the pressure sensors to the PIR sensors in acombined design whose overall sensing probability is shownin (6).)(Complement groupgroupgroupgroup UPUPUP  (6))( groupgroupgroup UPU  is the probability of the incorporation ofpressure sensors and PIR sensors to increase the sensingprobability. Table IV shows the overall probability by addingthe pressure sensors for sensing the intruder in the alert state.TABLE IVOVERALL SENSING PROBABILITIES AND INDIVIDUAL SENSINGPROBABILITYPgroup+Ugroup –( Pgroup ×Ugroup )Psingle Usingle 0.3 0.4 0.5 0.6 0.70.3 0.2253 0.3702 0.5599 0.7496 0.89450.4 0.3702 0.4880 0.6422 0.7964 0.91430.5 0.5599 0.6422 0.7500 0.8578 0.94010.6 0.7496 0.7964 0.8578 0.9191 0.96590.7 0.8945 0.9143 0.9401 0.9659 0.9856V. IMPLEMENTATION RESULTSTable V shows the sensors used in different operating states.We use PIR and pressure sensors as the sensors in the alertstate since these sensors consume little power. Because theultrasonic sensor has advantages in the sensing range, itconsumes, of course, more power. We use it and the PIRsensors in the detection state.TABLE VTHE SENSORS USED IN DIFFERENT OPERATING STATESSystem State Alert State Detection StatePIR sensor Use UsePressure sensor Use Not in useUltrasonicsensorNot in use UseTo enhance the sensing reliability, we use multiple PIRsensors and pressure sensors in alert sensor groups to detectan intruder. Each group uses the MVM management toimprove the sensing probability. We also install another alertsensor group indoors to avoid any miss-sensing outdoors. Fig.singleP)(nPmultiplen = 3 n = 5 n = 7 n = 9 n = … n ∞0.30 0.216 0.163 0.126 0.098 00.40 0.352 0.317 0.289 0.266 00.50 0.500 0.500 0.500 0.500 0.500 0.5000.60 0.648 0.682 0.710 0.733 1.0000.70 0.784 0.836 0.874 0.901 1.000
  5. 5. Y.-W. Bai et al.: Design and Implementation of a Home Embedded Surveillance System with Ultra-Low Alert Power 1579 shows that many alert sensor groups are placed both indoorsand outdoors in the alert state. When the alert sensor groupsenses an intruder, the MCU is woken up and turns on thepower supply for the detection sensor group.3583mmFig. 9. Our sensor placement in the alert state.Fig. 10 shows our sensor placement in the detection statewhere we use ultrasonic sensors and PIR sensors to sensean intruder. The ultrasonic transmission is blocked when anintruder enters the transmission path of the sensing area.But as the ultrasonic transmission spreads a beam anglewhich is likely to cause interference, we use multiplereceivers, and the MVM enhances the overall sensingprobability. The PIR sensor circuit has a least three groupsof PIR sensors placed on the ceiling of the specific roomenvironment.Fig. 10. Our sensor placement in the detection state.To find the average power consumption of the overalloperation we make Vdd the working voltage. IDetection is thecurrent value in the detection state. DutyCycleDetection is thedetection period, IAlert is the current value in the alert state andDutyCycleAlert is the alert period. The average powerconsumption is as shown in (7).Poweraverage = Vdd ×(IDetection×DutyCycleDetection)+(IAlert×DutyCycleAlert) (7)To reduce the average power consumption one can reducethe operating voltage. However, if we reduce the voltage ofthe sensor, the sensing probability of the PIR sensors may bereduced as shown in Fig. 11. It shows the sensing distance asabout 4 to 5 meters when the working voltage is 3V. Sincethe ceiling height is about 3 m, this is a suitable place for thePIR sensor group. Hence the PIR sensor can have a lowervoltage and still enough sensing probability. For the workingvoltage of the pressure sensors and the MCU we choose 3 Vfor our prototype.1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6405060708090100Sensing Distance (m)SensingProbability(%)3V4V5VFig. 11. The sensing probability of different working voltages.Table VI shows the power consumption of a single sensorgroup if there is no intruder. In this case the powerconsumption of a single PIR sensor group is about 1 mW,which increases to 7.3 mW when a PIR sensor detects anintruder. The pressure sensor consumes 0.8 mW, whichincreases to about 1.5 mW when it senses an intruder. Theultrasonic sensor group needs to constantly transmit andreceive signals, so it consumes around 256 mW.TABLE VITHE POWER CONSUMPTION OF A SINGLE SENSOR GROUPSystem StateAlert StatePower ConsumptionmWDetection StatePower ConsumptionmWPIR sensor 1.05 7.35Pressure sensor 0.8 1.5UltrasonicsensorNot in use 256.08We use the MCU which has a sleep mode to reduce thealert power, and we reduce the working clock of the MCU toits minimum clock rate to reduce the power consumption. At
  6. 6. 158 IEEE Transactions on Consumer Electronics, Vol. 57, No. 1, February 2011the same time the MCU maintains a speed faster than anintruder can move. Fig. 12 shows the measurement of thepower consumption for the system operation over some time.The top line shows the traditional surveillance system. Thelower line shows using the alert state.0 5 10 15 20 25 30 35 40 45 50051015Work Time (min)SystemPower(W)Our DesignPrevious DesignFig. 12. The measurement of the power consumption of the system inoperation.Table VII shows the comparison of the power consumptionbetween the operation states. This measurement shows thatour surveillance system with the alert state reduces theaverage power consumption.TABLE VIIA COMPARISON OF THE POWER CONSUMPTION BETWEEN THE OPERATIONSTATESPowerconsumptionin the alertstatePower consumptionin the detectionstateAverage powerconsumptionby staying90% in thealert state and10% in thedetection statePreviousdesign12 W 12 W 12 WOurdesign20 mW 11 W 1.102 WTable VIII shows the comparison between our previousdesign and other products. We have used a number ofindoor PIR and ultrasonic sensor groups in that design.Because continuing intrusion detection consumes morepower, which is around 12W, we have in this design addedPIR sensors and pressure sensors where the intruder mustpass through outdoors, thereby reducing the powerconsumption to 1.102 W from 12W by staying 90% in thealert state and 10% in the detection state, an improvementof 10.9 times.TABLE VIIICOMPARISON OF THIS AND OUR PREVIOUS DESIGN AND OTHER PRODUCTSPlatformOurdesignPreviousdesignProductAProduct BMiss rate Low Low Medium MediumVideo pixelsUp to1024*768Up to1024*768360*240 640*480Powerconsumptionin the alertstate20 mW 12 W 66 W 48 WPowerconsumptionin thedetectionstate12 W 12 W 66 W 48 WVolumeSOC(about10*10*3cm3)SOC (about10*10*3cm3)21*14*13(cm3)22*28*5 (cm3)Cost Low Low High HighVI. CONCLUSIONSIn this design we use multiple sensor groups with lowpower consumption for the detection of an intruder. The MCUstays in a sleep state, unlike the traditional surveillance systemwhich stays in the detection state. We reduce the powerconsumption in the alert or sleep state by 10.9 times byremaining 90% in the alert state and 10% in the detection state,and we use two sensor groups to improve the detectionreliability of the alert state. In addition our home embeddedsurveillance system reduces unnecessary memoryconsumption for the capture of images without an intruder,compared to previous surveillance systems.REFERENCES[1] Cheng-Hung Tsai, Ying-Wen Bai, Wang Hao-Yuan and Ming-Bo Lin,“Design and Implementation of a Socket with Low Standby Power”,IEEE Transactions on Consumer Electronics, Vol. 55, No. 3, pp. 1558-1565, August 2009.[2] A. Meier and W. Huber, “Results from the investigations on leakingelectricity in the USA,” Lawrence Berkeley National Laboratory,California, 1998.[3] J. P. Ross and A. Meier, “Measurements of whole-house standby powerconsumption in California homes,” Energy, vol. 27, pp. 861-868, Sep.2000.[4] A. Meier, “A worldwide review of standby power use in homes,”Lawrence Berkeley National Laboratory, Dec. 2001.[5] K. Clement, I. Pardon, and J. Driesen, “Standby Power Consumption inBelgium,” in Proc. EPQU, pp. 1-4, Oct., 2007.[6] International Energy Agency, Things That Go Blip in the Night: StandbyPower and How to Limit It, Paris, France, International Energy Agency,2001.[7] International Energy Agency, Standby Power Use and the IEA “1-wattPlan”, International Energy Agency, April 2007.[8] L. McGarry, “The standby power challenge,” IEEE Transactions onAsian Green Electronics, pp. 56-62, 2004.[9] Bo-Teng Huang, Ko-Yen Lee and Yen-Shin Lai, “Design of a Two-Stage AC/DC Converter with Standby Power Losses Less than 1 W,”Proc. PCC, pp. 1630-1635, Apr. 2007.[10] Jee-Hoon Jung, Jong-Moon Choi and Joong-Gi Kwon, “Noveltechniques of the reduction of standby power consumption for multipleoutput converters,” Proc. APEC, pp. 1575-1581, Feb. 2008.[11] S. Y. R. Hui, H. S. H. Chung, and D. Y. Qiu, “Effective standby powerreduction using non-dissipative single-sensor method,” Proc. PESC, pp.15-19, Jun. 2008.
  7. 7. Y.-W. Bai et al.: Design and Implementation of a Home Embedded Surveillance System with Ultra-Low Alert Power 159[12] Yu-Kang Lo, Shang-Chin Yen, and Chung-Yi Lin, “A High-EfficiencyAC-to-DC Adaptor with a Low Standby Power Consumption,” IEEETransactions on Industrial Electronics, pp. 963-965, Feb. 2008.[13] Chia-Hung Lien, Chi-Hsiung Lin, Ying-Wen Bai, Ming-Fong Liu, andMing-Bo Lin, “Remotely Controllable Outlet System for Home PowerManagement,” Proceedings of 2006 IEEE Tenth InternationalSymposium on Consumer Electronics (ISCE 2006), St. Petersburg,Russia, pp. 7-12, June 28-July 1, 2006.[14] Chia-Hung Lien, Ying-Wen Bai, and Ming-Bo Lin, “Remote-Controllable Power Outlet System for Home Power Management,” IEEETransactions on Consumer Electronics, pp. 1634-1641, Nov. 2007.[15] Ying-Wen Bai, and Yi-Te Ku, “Automatic room light intensity detectionand control using a microprocessor and light sensors,” IEEETransactions on Consumer Electronics, pp. 1173-1176, Aug. 2008.[16] Ying-Wen Bai, Zong-Han Li and Zi-Li Xie, “Enhancement of theComplement of an Embedded Surveillance System with PIR Sensors andUltrasonic Sensors”, The 14th IEEE International Symposium onConsumer Electronics, Braunschweig, Germany, pp. 1-6, June 07-10,2010.[17] Ying-Wen Bai and Jung-Lu Chen, “The Enhancement of SpeechRecognition Probability by Homogeneous Majority Voting Mechanism,”The International Association of Science and Technology forDevelopment, Signals, and System, pp.48-53, Nov. 2004.[18] Jie Cao and Li Li, “Vehicle Objects Detection of Video Images Based onGray-Scale Characteristics,” ETCS 09, First International Workshop onEducation Technology and Computer Science, 2009, pp.936-940, 7-8Mar. 2009.[19] Francesco Alonge, Marco Branciforte and Francesco Motta, “a novelmethod of distance measurement based on pulse position modulation andsynchronization of chaotic signals using ultrasonic radar systems,” IEEETransactions on Instrumentation and Measurement, Feb. 2009, pp. 318-329.[20] SeungKeun Cho, TaeKyung Yang, MunGyu Choi and JangMyung Lee,“Localization of a high-speed mobile robot using global features,” 4thInternational Conference Autonomous Robots and Agents, ICARA 2009,pp.138-142, 10-12 Feb. 2009.BIOGRAPHIESYing-Wen Bai is a professor in the Department ofElectrical Engineering and Graduate Institute of AppliedScience and Engineering at Fu-Jen Catholic University.His research focuses on mobile computing andmicrocomputer system design. Ying-Wen Bai obtainedhis M.S. and Ph.D. degrees in electrical engineering fromColumbia University, New York, in 1991 and 1993,respectively. Between 1993 and 1995, he worked at theInstitute for Information Industry, Taiwan.Zi-Li Xie is currently working toward the M.S. degree inElectronic Engineering at Fu-Jen Catholic University,Taiwan. His research focuses on video streaming forembedded surveillance systems. He received his in electronic engineering from Lunghwa Universityof Science and Technology in 2007.Zong-Han Li is currently working toward the M.S. degreein Electronic Engineering at Fu-Jen Catholic University,Taiwan. His research focuses on video streaming forembedded surveillance systems. He received his in Computer Science and Information Engineeringfrom China University of Technology in 2007.