適用於大範圍的合作式定位方法 - 政大學術集成

全文

(1)國立政治大學資訊科學系 Department of Computer Science National Chengchi University 碩士論文 Master’s Thesis. 立. 政治大. ‧ 國. 學 ‧. 適用於大範圍的合作式定位方法. n. al. er. io. sit. y. Nat. A large scale cooperative localization method. Ch. engchi. i n U. v. 研究生：鄧偉敦指導教授：蔡子傑. 中華民國一百年六月 June 2011.

(2) 適用於大範圍的合作式定位方法 A large scale cooperative localization method. 研究生：鄧偉敦. Student：Wei-Tun Teng. 指導教授：蔡子傑. 政治大國立政治大學資訊科學系碩士論文. 學. ‧ 國. 立. Advisor：Tzu-Chieh Tsai. y. ‧. Nat. sit. A Thesis. er. io. submitted to Department of Computer Science. n. al National Chengchi University iv. n U i Requirements in partial fulfillment e n g cofhthe. Ch. for the degree of Master in Computer Science. 中華民國一百年六月 June 2011.

(3) 適用於大範圍的合作式定位方法摘要. 近來幾年智慧型手機和適地性服務已經變得非常熱門。智慧型手機必須有辦法知道使用者的位置，因此精準的定位技術就變得重要。. 政治大. 至今人們只能夠在特定的環境利用某些定位方法，就像是 GPS 只能. 立. 用於室外的空間。但是人們總是生活於大範圍的環境像是校園、都. ‧ 國. 學. 市和觀光區，這種大範圍的環境包含了室內和室外的空間。. ‧. 大範圍的環境下，單一定位技術未必到處都可用，因此我們結合. y. Nat. er. io. sit. 了 GPS、WiFi 和物聯網定位提出了一個異質式定位演算法。我們. n. 提出“定位可能性”來選擇比較“可靠”(可能)的定位方法。除此 a v. i l C n h e n g c h i U WiFi 訊號強度可更進之外，利用較可靠的鄰近使用者與自己之間的一步改善定位的精準度。特別針對某些使用者在沒有任何可用的定位方法時更有幫助。這個方法被稱為“合作式定位”。. 最後，我們用模擬來評估我們演算法的精準度。因為訊號強度每分每秒都在波動，因此我們測量實際的訊號強度和 GPS 放入模擬器，讓實驗結果變得更真實。最後我們也證明我們的演算法可以做在手機上而且更精準。 i.

(4) A large scale cooperative localization method Abstract Smart phones and Location Based Services (LBSs) have become very popular in recent years. To this end, the smart phone needs to know the locations of users. Therefore, an accurate localization technique is important. To date people can use some localization systems in some specific areas. For instance, GPS can only be used in the outdoor space.. 政治大. However, people always live in large scale environments like campus,. 立. urban and tourist areas. The large scale environments should include. ‧ 國. 學. indoor and outdoor space.. For large scale environment, a single location technique is not always. ‧. available everywhere. Therefore, we proposed a heterogeneous. Nat. sit. y. localization algorithm which combines GPS, WiFi and Internet Of Things. er. io. (IOT) localizations. We proposed ―localization possibility‖ for each. n. a l algorithm use localization localization methods. This i v possibility to select. n U e nBesides, the most ―reliable‖ (possible) one. g c h ithe more reliable nearby users. Ch. can further enhance the localization by measuring the relative WiFi signal strength. It helps especially for those users who have no any available localization methods. This method is called ―cooperative localization‖. Finally, we evaluated the accuracy of our algorithms by simulation. Because signal strength fluctuates from minute to minute, we measured empirical data and put into the simulator to make our experimental results more real. Finally, we also verify that our idea can be implemented on smart phones and our algorithm is more accurate. ii.

(5) 致謝辭. 這本論文得以完成首先要感謝我的指導教授，蔡子傑老師。在碩一接下展場定位的計畫後就積極的督促我，碩二時不厭其煩的與我討論這篇碩士論文的內容，在口試前一次次的陪我 dry run，和口試時的協助，感謝老師細心的指導。另外要感謝口試委員，吳曉光、周承復、陳伶志及林宗男等老師，百忙中抽空前來，在口試時提出寶貴的建議，使這份論文更加完整。. 政治大. 然後我要感謝在政大資科兩年內認識的同學，謝謝文卿、貞慈學姊，育晟、. 立. 勇麟、界誠、泰榮、諭祺學長，在學業、就業和生活上的幫忙和指點，特別謝謝. ‧ 國. 學. 界誠學長總是陪我去重訓室，讓我不至於從早到晚都苦悶的待在實驗室，謝謝惠翔在程式撰寫上的提點，謝謝我的同學凱禎、彥嵩、智杰、筱慈、建彪、佩璇、. ‧. 鐘毅和思釆在修課和研究上的互相幫忙，還有謝謝和我一起口試的在職專班同學. Nat. sit. y. 志宏，總是提醒我沒注意到的細節，一起準備口試餐點，一起討論論文的寫作，. n. al. er. io. 謝謝和我一起在社科院打工的奕愷，謝謝學弟欣諦、英明、昶瑞、冠傑和郁翔在. i n U. v. 實驗上的幫忙，另外感謝政大服務科學中心的劉佩雯小姐幫我解決很多帳務上的. Ch. engchi. 問題，讓我能有更多的時間在做研究上，最後感謝我認識的所有資科和數位內容的同學，祝各位學業順利，身體健康。在我寫程式的過程中要感謝我的電腦 iMac 和 x200 都沒有發生資料遺失的問題。最後我要感謝我的父母，給我這個機會讓我衣食無缺的一路唸到碩士，以及平時給我諸多的鼓勵、幫助和包容。. iii.

(6) TABLE OF CONTENTS. CHAPTER 1 Introduction ..........................................................................1 1.1 Background ....................................................................................1 1.2 Motivation ......................................................................................2 1.3 Organization ...................................................................................3 CHAPTER 2 Related Work ........................................................................4 2.1 Collaborative localization [16] ......................................................6. 政治大. 2.2 Virtual Compass [17] .....................................................................8. 立. CHAPTER 3 Localization Algorithm .......................................................10. ‧ 國. 學. 3.1 WiFi localization ..........................................................................10. ‧. 3.2 Heterogeneous localization ..........................................................16 3.3 Cooperative localization ..............................................................22. y. Nat. io. sit. CHAPTER 4 Experimental Evaluation ....................................................25. er. 4.1 Simulation setup...........................................................................25. n. a. v. 4.2 Simulation resultl.......................................................................... 31 ni C. hengchi U. 4.3 Implementation setup ...................................................................36 4.4 Implementation result ..................................................................37 CHAPTER5 Conclusions .........................................................................42 Reference ..................................................................................................43. iv.

(7) LIST OF FIGURE Figure 1: Outline of current wireless positioning systems [15]. .................... 6 Figure 2: An example of filter. ....................................................................... 7 Figure 3: An example of spatial placement for Virtual Compass. ................. 9 Figure 4: ORBI positioning system [18]. ..................................................... 11 Figure 5: ORBI navigation device [19]. ...................................................... 12 Figure 6: ORBI navigation device [19]. ...................................................... 12. 政治大 Figure 8: The flowchart of WiFi localization system. ................................. 13 立 Figure 7: Cumulative percentage of error distance in Cycle show. ............. 12. ‧ 國. 學. Figure 9: Data collection in the exhibition. ................................................. 14 Figure 10: Example for the intersection of two normal distributions of one. ‧. access point [20]. ................................................................................. 15. sit. y. Nat. Figure 11: An example of heterogeneous localization. ................................ 18. io. er. Figure 12: Flow chart of heterogeneous localization. .................................. 19. al. Figure 13: Two human movement capacities are not intersection. .............. 19. n. v i n Cofhmoving path. ...................................................... Figure 14: An example 20 engchi U Figure 15: An example for possibility reduction. ........................................ 21 Figure 16: Two HTC Desire smart phones. (right opens hotspot mode) .. 22 Figure 17: An example for cooperative localization. ................................... 23 Figure 18: Flow chart of cooperative localization. ...................................... 24 Figure 19: Flow chart of simulator............................................................... 26 Figure 20: An example of site survey. ......................................................... 27 Figure 21: An example of path..................................................................... 27 Figure 22: Distance (Meter) to WiFi signal strength (RSSI) in outdoor space. v.

(8) X axis is distance between two Desire and Y axis is relative WiFi signal strength. ..................................................................................... 28 Figure 23: Distance(Meter) to WiFi signal strength(RSSI) in indoor to outdoor space. ...................................................................................... 29 Figure 24: Distance(Meter) to WiFi signal strength(RSSI) in indoor space. .............................................................................................................. 29 Figure 25: WiFi localization possibility and error distance in all paths. ..... 30 Figure 26: Four paths in simulation. ............................................................ 31. 政治大 Figure 28: Simulation result in path 2.......................................................... 33 立 Figure 27: Simulation result in path 1.......................................................... 32. Figure 29: Simulation result in path 3.......................................................... 34. ‧ 國. 學. Figure 30: Simulation result in path 4.......................................................... 35. ‧. Figure 31: A picture of our experimental. .................................................... 37. sit. y. Nat. Figure 32: First implementation result in outdoor path. .............................. 38. io. er. Figure 33: Second implementation result in outdoor path. .......................... 39 Figure 34: Implementation result in indoor path. ........................................ 40. n. al. Ch. engchi. vi. i n U. v.

(9) CHAPTER 1 Introduction 1.1 Background. 政治大. Personal Digital Assistant (PDA) and smart phone have become very powerful, i.e. Android. 立. These devices always provide wireless internet service and Global. Positioning System (GPS).. 學. ‧ 國. phone and iphone.. These developments have fostered the rise of a novel class of. applications called Localization based Services (LBSs).. LBSs provide added value by. ‧. Nat. y. enabling services such Resource tracking, Finding someone and something and. rental equipment, doctors, fleet scheduling.. Finding someone or something is like Person by. er. n. al. sit. Resource tracking is like taxis, service people,. io. proximity-based notification (push or pull).. i n U. v. skill (doctor), business directory, navigation, weather, traffic, room schedules, stolen phone, emergency calls.. Ch. engchi. Proximity-based notification is like Targeted advertising, buddy list,. common profile matching (dating), automatic airport check-in.. So an accurate localization. method is important. For outdoor localization systems, GPS is easy to use and offers a good localization accuracy and reliability.. But the mobile need line of sight to at least three GPS satellites.. So it is inefficient in indoor environments and its accuracy degrades in urban environments. On the other hand, the energy consumed by GPS device is a significant deterrent. Indoor localization technique has become more popular in recent years.. It can consider. the location detection of products stored in a warehouse, location detection of medical 1.

(10) personnel or equipment in a hospital, localization detection of firemen in a building on fire, detecting the location of police dogs trained to find explosives in a building, and finding tagged maintenance tools and equipment scattered all over a plant.. Radio Frequency. Identification (RFID) and WiFi localization can be used in indoor space. good accurate localization.. RFID can provide. SpotON[1] and LANDMARC[2] are RFID localization systems.. These systems have been proposed during the last few years. hardware like RFID readers and tags. fingerprinting method in this paper.. 立. But it requires specialized. WiFi localization is to belong to location. 治政 It using existing wireless access points for indoor 大. localization as an emerging technique has been widely studied and deployed, because WiFi. ‧ 國. 學. access points are provided in many environments e.g. campus, office, convenience store and Many smart phones also provide WiFi to connect to internet.. WiFi localization. ‧. uses a two phase approach.. During offline phase, the location coordinates/signal strengths. Nat. During the online phase, a localization technique. sit. from nearby base stations are collected.. y. restaurants.. al. n. an estimation location.. er. io. uses the currently observed signal strengths and previously collected information to figure out. i n U. v. Microsoft RADAR [5] proposed this approach.. Ch. engchi. But the WiFi. localization needs for very time consuming creation of the fingerprint database and it need a lot of nearby wireless access points. Therefore, WiFi localization is not applicable in large outdoor space.. 1.2 Motivation There are indoor and outdoor spaces in urban environments or on campus. always staying, walking or biking in this environment.. But GPS, WiFi and RFID. localization have some drawback in indoor and outdoor environment. indoors and its accuracy degrades in urban environments. 2. Most people are. GPS rarely works. RFID localization needs a lot of.

(11) active RFID tags and RFID readers.. WiFi localization can be used indoor and outdoor. environments but it needs a lot of time and people to create fingerprint database before we estimate our locations. So we proposed a localization method in large scale environments.. Our method uses. WiFi localization in outdoor and indoor space and uses GPS in outdoor space.. Sometime. GPS signal strength is weak in somewhere then we can place more WiFi access points and collect more WiFi signal strength during offline phase. We also consider that there are many. 政治大. RFID readers in urban environments and on campus like MRT payment systems and RFID. 立. People can use RFID and Near Field Communication. (NFC) device to get reader’s locations. like GPS, WiFi and NFC.. 學. ‧ 國. access control systems on campus.. Our method combines the device on smart phone. We want people need not take another localization device.. ‧. Furthermore a lot of people use iphone or Android phone and they open WiFi to connect to. y. Nat. io. have GPS but many people open GPS around the user.. n. al. sit. Or there are no WiFi access points and infrastructures around us and user doesn’t So user can scan nearby user’s. er. internet.. i n U. v. relative WiFi signal strength. Using nearby user’s location and relative signal strength to estimate user’s location.. Ch. engchi. Finally we implement the localization system in Android phone. and experiment on down-hill campus of NCCU.. 1.3 Organization The rest of this thesis is organized as follows.. Chapter 2 introduces related works in WiFi. localization systems and other localization methods. algorithm of our localization algorithm. implementation results.. Chapter 3 introduces the basic idea and. Chapter 4 presents simulation results and. Finally, Chapter 5 concludes this paper and gives some directions. for future work. 3.

(12) CHAPTER 2 Related Work GPS and GSM 政治大 GPS 立performs well in outdoor open –sky environments but tall. Localization techniques can be classified to Indoor and outdoor localization. is outdoor localization.. ‧ 國. 學. buildings prevent users from getting a line-of-sight to GPS satellites. degrades in urban environments.. GSM localization is widely available but it provides poor. ‧. accuracy without a fingerprint profile, or outside city center.. UWB localization is triangulation. n. er. io. technologies like Ultra-wideband (UWB), RFID, WLAN.. It always uses wireless. sit. y. Nat. Indoor localization has become very popular in recent ten years.. al. So its accuracy. v Triangulation uses the i n. like TOA(time of arrival), TDOA(time difference of arrival).. Ch. engchi U. geometric properties of triangles to estimate the target location and it need line of sight between transmitter and receiver.. Therefore, obstacle will cause inaccurate.. UWB’s transmission distance is too short.. So we need use a lot of UWB tags.. Furthermore RFID. localization like LANDMARC [2] and it needs to place a lot of RFID readers and active tags on the localization area.. Tags can measure reader’s signal strength.. And then compare the. signal strength between user’s tag and the tags we placed. WLAN (include WiFi) localization method is more popular for indoor localization and is also called location fingerprinting method. offline phase and online phase.. There are two phase for location fingerprinting:. During the offline phase, a site survey is performed in an 4.

(13) environment.. The location coordinates/labels and respective signal strengths from nearby. base stations (access point)/measuring units are collected.. During the online phase, a. localization technique uses the currently measured signal strengths and previously collected information to figure out an estimated location.. The main challenge to the techniques based. on location fingerprinting is that the received signal strength would fluctuate. proposed the location fingerprinting algorithm. deterministic ones and probabilistic ones.. Many people. They are classed as two categories:. Deterministic approaches like early system. 政治大. RADAR [3]. It calculates the Euclidean distance in signal space to compare the signal. 立. Ekahau [5] and others [6, 7].. Probabilistic approaches like Horus [4],. 學. ‧ 國. strength between online phase and offline phase.. It calculates signal strength distribution during offline phase. and it computes a probability for each fingerprint at every coordinate in site survey during. ‧. It offers a better robustness with respect to noisy signals.. Furthermore, a lot. Nat. y. online phase.. n. al. er. And some people research how to place AP to improve accuracy [10], but. io. fluctuation [8, 9].. sit. of algorithms want to use history localization information to reduce effect of signal strength. i n U. v. we can’t always control to deployment APs in large scale environments.. Ch. engchi. WLAN localization. not needs specialized hardware, because there are dense WiFi access points on campus and at urban environments and to date many smart phones offer WiFi.. And WLAN’s transmission distance is. time-consuming deployment of infrastructure. around 50~100 meter longer than UWB.. So it avoids expensive and. But these systems typically require a. time-consuming and costly build the radio map during offline phase. Therefore, there are some papers propose to let users create radio map if user know where is him [11, 12]. And some papers analyze location fingerprinting algorithm and want to improve the accuracy by filter algorithms or sensor fusion [13, 14]. Figure 1 depicts a rough outline of the current wireless-based localization system [15]. 5.

(14) It shows the scale and resolution for the localization systems.. UWB can use in indoor and. outdoor environments but it is applicable only for line-of-sight environments. RF and ultrasonic hybrid methods are used only in indoor environments. localization perform well in indoor and outdoor locally environments.. RF and IR,. WLAN. Therefore, we want to. use GPS and WLAN to propose heterogeneous localization in large scale environments. large scale environments include indoor and outdoor environments.. The. We introduce some. exiting WiFi based localization systems and analyze their drawback in large scale environments.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 1: Outline of current wireless positioning systems [15].. 2.1 Collaborative localization [16] Chan et al. proposed a confidence model to enhance position estimation by leveraging more 6.

(15) accurate location information from nearby neighbors within the same cluster.. They observed. the positional accuracy of WiFi-based positioning engines is influenced by people clustering due to electromagnetic interference.. In these systems, the position estimated by RSSI. fingerprint usually feeds into a particle filter to constrain location estimation within a reasonable variation consistent with human movement. This filter considers human movement capacity.. Figure 2 is an example for filter.. For example people usually can’t move. exceed three meter per second so his human movement capacity is three meter per second.. 政治大. Point at time t is the previous location estimation for time t+1.. 立. Then blue point(estimation. from location engine) is location estimation at time t+1 by WiFi localization engine, but. blue point to the human movement capacity bound.. Nat. The sum of dt+1, dt+2 and dt+3 is in inverse proportion to. n. al. er. io. confidence score.. And dt+1, dt+2 and dt+3 is the correcting. sit. distance at t+1, t+2 and t+3.. So the red point(estimation from. y. particle filter) at t+1 is the estimation from filter.. And filter corrects the. ‧. ‧ 國. 學. people can’t move so far from red point at time t in time interval.. Ch. engchi. i n U. v. Figure 2: An example of filter.. 7.

(16) They found confidence in location estimation correlates highly to positioning stability of a target node computed over time from a particle filter.. By finding nearby targets with. proximity sensors (Zigbee), nodes with lower confidence could improve their estimation accuracy by leveraging more accurate location information from nearby neighbors within the same cluster.. This collaborative localization has two drawbacks in large scale environments.. First, they use sensor to detect nearby people within 2 meter. for large scale environments.. We think 2 meter is too short. Second, the system is only WiFi localization system.. 政治大. It is. only used in indoor and outdoor locally environments.. 學. ‧ 國. 立 2.2 Virtual Compass [17]. Virtual Compass is a peer-based localization system for mobile phones.. It does not require. ‧. Nat. n. al. er. Where node A is calculating a placement for itself with respect to 3. io. creating neighbor graph.. Figure 3 is an example for Virtual Compass. sit. and Bluetooth to create a neighbor graph.. y. any infrastructure support, but instead uses multiple, common radio technologies like WiFi. other nodes, and begins by placing itself at the origin.. Ch. i n U. v. It finds the peer, B, that is the. e n g c h i Since our underlying distance estimation. smallest distance (r1) away, and places it at (0, r1).. is unable to discern direction, this placement is arbitrary as long as it is r1 distance from the origin.. Next, they choose node C because it is constrained by both A and B.. There are. multiple solutions for placing C since the constraints are quadratic, and they randomly choose one.. Next, they again choose another node D that is constrained by as many of the currently. placed nodes as possible.. Again, there are multiple solutions and they select the set of. locations that is the closest together and take an average between these coordinates to place D. They run this algorithm multiple times with different constraint orderings and they use average of the coordinates. 8.

(17) 政治大. 立. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 3: An example of spatial placement for Virtual Compass.. They found the average error in spatial placement of nine nodes in a 100m2 area was 1.9 meters.. Virtual Compass is focus on relative positioning, because mobile social applications. are heavily driven by the relative positioning of people. need to use absolute location.. But in large scale environment we. In figure 3, if node A and B use GPS and get locations then. these nodes can locate other node.. But it doesn’t consider who is reliable? An inaccurate. location will cause neighbors’ location error. 9.

(18) CHAPTER 3 Localization Algorithm We want to locate people in large scale environment, but we can’t use only one localization algorithm of related work.. 政治大. Now, we enhanced the filter and combined GPS, WiFi and. 立. localization offers good accuracy in indoor environment. outdoor environment where deployment a lot of APs.. WiFi. 學. ‧ 國. IOT(internet of things) localization to propose a heterogeneous localization.. And sometime it can use in. On the other hand, we implemented a. ‧. WiFi localization system at Taipei World Trade Center Nangang Exhibition Hall from July,. y. Nat. n. al. er. Therefore, we embed this algorithm [7] in our method. Furthermore we. io. different shows.. sit. 2009. And the average accuracy of this system is 4.2 meter in many experimental at. i n U. v. consider nearby neighbor’s relative WiFi signal strength and their locations to correct the. Ch. engchi. location estimation, and this method called cooperative localization. introduce WiFi localization system using probabilistic approach. heterogeneous localization.. In section 3.1, we. In section 3.2, we propose. In section 3.3, we consider nearby neighbors to propose. cooperative localization.. 3.1 WiFi localization Service Science Research Center(SSRC) in NCCU with us want to development an WiFi localization system at Taipei World Trade Center Nangang Exhibition Hall to provide smart navigation for foreign buyers.. Figure 4 is that navigation system. 10. Figure5, 6 are the.

(19) navigation device.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 4: ORBI positioning system [18]. 11.

(20) Figure 5: ORBI navigation device [19].. Figure 6: ORBI navigation device [19].. 政治大 Machine Tool Show and Cycle show. The average accuracy is 4.2 meter in Taipei Cycle 立. And we tested our localization system in 2010 Taipei Int'l Electronics Show、2011 Taipei Int'l. ‧. ‧ 國. Figure 7 is the experimental result in the Cycle show.. 學. io. sit. y. Nat. n. al. er. show.. Ch. engchi. i n U. v. Figure 7: Cumulative percentage of error distance in Cycle show. 12.

(21) Now we present the WiFi localization system.. Figure 8 shows the flowchart of the. system.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 8: The flowchart of WiFi localization system.. 13.

(22) 政治大. 立. Figure 9: Data collection in the exhibition. Pink fill grids are radio map.. ‧ 國. 學. 1.. Data Collection:. ‧. Nat. And we measured 20 times at one point by Android device.. sit. signal strengths every 6~8 meter.. y. There are 56 APs every 17 meter in exhibition. During the offline phase, we measured. n. al. er. io. Then we saved these coordinates and signal strengths to database.. i n U. v. every show, because decorations are not the same in different show.. Ch. engchi. cost an afternoon in Nangang exhibition. 2.. We collect data before And ten people will. Radio Map Creation:. Our system load those raw data at one point in database and calculated mean and standard deviation of every AP.. This called radio map and the point in radio map called radio map. point. 3.. Tracking Data Collection:. During online phase, user take Android device to measure signal strength, and the user called tracking point.. Every tracking points measure signal strength five times and the system. calculate mean and standard deviation of every AP. 14.

(23) 4.. Target Determination:. The system calculated the similarity [7] between tracking point and every radio map points to search five closest match radio map points.. These five points called targets.. The algorithm. creates two signal strength normal distributions to tracking point and radio map for one access point and computes the intersection area of the two density distributions.. Figure 10 is an. example for the intersection of two normal distributions of one access point.. 政治大. TP. RMP. 學. ‧ 國. 立. RMP. TP. ‧. io. sit. y. Nat. n. al. er. Figure 10: Example for the intersection of two normal distributions of one access point [20]. (good match on the left, bad match on the right hand site). Ch. engchi. i n U. v. +∞ 𝑃 Overlap(TP, RMP) = ∫−∞ (𝑚𝑖𝑛 { 𝑇𝑃 ) 𝑃 𝑅𝑀𝑃. (1). Similarity(TP, RMP) = ∑𝑛𝑖=1 𝑂𝑣𝑒𝑟𝑙𝑎𝑝(𝑇𝑃𝑖, 𝑅𝑀𝑃𝑖)/m. (2). Equation 1 calculates the overlap area between tracking point and a radio map point of one access point.. TP is tracking point.. RMP is one of radio map point.. And Equation 2 is the. average overlap of all access points between tracking point and one radio map point. set of common access points. m is the number of all access points contained in either. 15. n is the.

(24) 5.. Interpolation and Location Estimation:. We use five target points and their weight to interpolate our location estimation.. The weight. of every target points is the inverse of Euclidean distance between tracking point and the target point in signal space. ̅̅̅̅ - 𝑇𝑎𝑟𝑔𝑒𝑡 ̅̅̅̅̅̅̅̅̅̅ || W = 1/||𝑇𝑃 i. (3). i 2. ̅̅̅̅ here is the vector of the means of the RSSI values of each Equation 3 displays the weight, TP ̅̅̅̅̅̅̅̅ is the vector of the means of the RSSI values of each AP at the ith target. AP, Target. 政治大 then we calculate weighted sum by targets’ coordinate and weight. 立 i. ‧ 國. Filtering:. 學. 6.. And. Every APs have signal strength fluctuations from minute to minute, and the fluctuations arise. ‧. from motion people.. It course the predicted location is far away from the previous location,. y. sit. io. n. al. er. in related work.. Nat. Therefore we use filter to correct this location estimation like Collaborative localization [16]. Ch 3.2 Heterogeneous localization e. ngchi. i n U. v. We use WiFi localization in Nangang exhibition, because there are a lot of APs.. But in this. paper we focus on the large scale environment, and AP is sparse in more outdoor space. the other hand, there are a lot of RFID readers in urban environments and on campus.. On And. we can use NFC or RFID tags in the smart phone to get the location from those readers. Therefore, we proposed a heterogeneous localization algorithm to combine GPS, IOT and WiFi localizations.. If we can get two or three locations at the same time, and which location. is accuracy and reliability?. We use localization possibility to solve this problem.. higher possibility express the localization is more accuracy. 16. The. Our method gives IOT.

(25) localization possibility is 1 because IOT reader and tag communication distance is very short so the location is reliable.. If some GPS devices can get its localization reliability in different. area and GPS localization possibility is set to the reliability.. But default GPS localization. possibility is 0.35, because based on our experience GPS localization possibility is 0.35 on NCCU campus and we will explain in our simulation.. GPS localization possibility is not 1,. because GPS is not always accurate in outdoor space.. WiFi possibility is given by first. target point in WiFi localization algorithm.. Localization possibility is 0 when the. 治政 Localization 大possibility will be reduced when. localization method can’t estimate location.. 立. user move in localization time interval because if user can’t estimate his location at a long. ‧ 國. 學. time and he can use previous localization coordinate but its reliability should be low. we use the localization possibility to enhance the filter.. And. The filter corrects the location when. ‧. the location estimation exceeds the human movement capacity.. Nat. y. But when the previous. n. al. So our method compares the possibility. er. This causes error localization estimation.. io. capacity.. sit. location estimation is not accurate then the new location is bounded in the human movement. i n U. v. between previous location estimation and new location estimation.. Ch. engchi. If the possibility of new. location estimation is higher than the previous one and the location estimation will jump. Figure 11 is an example for heterogeneous localization. heterogeneous localization.. Figure 12 is the flow chart of. We explain our algorithm by figure 11 and 12.. estimated on Lt by localization system at time t. system use GPS WiFi and IOT localization.. And then user move.. Now user is. The localization. And they have their location possibility.. GPS’s possibility is given by experience and we will expand that how we set the possibility in simulation.. WiFi localization possibility is the similarity of first target in WiFi localization.. IOT’s possibility is 1 when it gets reader’s location.. We select the coordinate which have. highest possibility at time t+1 and the coordinate is L̃t+1. 17. At the same time system reduces Pt.

(26) because user moves from time t to t+1. and Pt’ is 0.9.. Pt’ is the localization possibility that reduce from Pt. Then we calculate moving speed by localization history.. Assume the system. refer previous 4 times localizations and calculate the average moving speed.. Therefore, we. get human movement capacity by multiplying moving speed and time interval from time t to t+1.. We paint two human movement capacities around Lt and 𝐿̃t+1.. If there are. intersection with two human movement capacities and the localization estimation is on the Lt+1 is the localization result.. and 𝑃̃t+1.. 立. And Pt+1 inherit the higher possibility of Pt’. 政治大. ‧. ‧ 國. 學. io. sit. y. Nat. n. al. er. middle of them.. Ch. engchi. i n U. v. Figure 11: An example of heterogeneous localization.. 18.

(27) 立. 政治大. ‧ 國. 學. Figure 12: Flow chart of heterogeneous localization.. ‧. If there is no intersection of two human movement capacities like figure 13.. In case A, Pt’ is. y. sit. In case B, If Pt’ is lower than ̃ Pt+1 and we think L̃t+1 is more accurate.. io. er. capacity around Lt.. Nat. ̃t+1 and we use filter to correct the location estimation to human movement higher than P. So location estimation is on the human movement capacity around L̃t+1.. n. al. Ch. engchi. i n U. v. Case A.. Case B.. Figure 13: Two human movement capacities are not intersection. Case A is using filter. Case B is jumping to new location.. 19.

(28) In heterogeneous localization, we used localization possibility to select a localization method from GPS, WiFi and IOT localization and we didn’t use three localization coordinates simultaneously.. Because we think some localization is not accurate.. Localization accuracy. will be down if heterogeneous localization used these inaccurate localization results. In heterogeneous localization we reduce localization possibility of previous localization estimation.. We consider the possibility reduction is respect to moving path like figure 14.. 立. 政治大. ‧. ‧ 國. 學. Path 1:. Path 2:. n. al. er. io. sit. y. Nat Path 3:. Ch. engchi. i n U. v. Figure 14: An example of moving path.. If user moves in path 1 and path 2 by the same speed and time. reduction in path 1 is smaller than path 2. than path 2.. Because the moving distance in path 1 is shorter. And furthermore we consider path 3.. localization possibility shouldn’t reduce.. We think the possibility. The user doesn’t move in path 3 and the. Figure 15 is an example for possibility reduction.. When user moves to Lt+1 from Lt at time t+1 and the possibility of Lt should be reduced. refer the localization estimation from Lt-3 to Lt. Lt, Lt-1, Lt-2 and Lt-3.. We. First we calculate mean coordinate Lmean of. Second calculate the distance dt, dt-1, dt-2 and dt-3 from Lt, Lt-1, Lt-2 and 20.

(29) Lt-3 to Lmean.. Third we calculate the variance of dt, dt-1, dt-2 and dt-3.. 4 to calculate possibility reduction.. We consider variance of previous localization and move. time interval to propose the equation 4. previous localization possibility. from t to t+1.. Forth we use equation. Pt+1 is localization possibility at time t+1.. Pt is. σ is variance of dt, dt-1, dt-2 and dt-3. ∆t is time interval. D and T are constant.. We set D is the distance with user general moving. distance in four localizations and set T is general time interval in four localizations. Localization interval time in our system is constant.. The possibility is 0<P<=1, so we use. 政治大. minima with 1 and maxima with 0 in equation 4.. 立. 𝜎. ∆t. Pt+1 = Pt − min(𝐷 , 1) × min( T , 1). (4). ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 15: An example for possibility reduction.. 21.

(30) 3.3 Cooperative localization We proposed a heterogeneous localization algorithm to combine GPS, WiFi and IOT localizations in large scale environments. localization, how do I get my location?. And furthermore if I can’t use GPS, WiFi and IOT So we consider nearby neighbors’ relative WiFi. signal strength and location to proposed cooperative localization method. problems in cooperative localization. between two smart phones?. There two. First, how can we scan relative WiFi signal strength. Second, how can we deliver the location to my neighbors.. 政治大 One smart phone open WiFi. We. use hotspot mode to simulate ad-hoc mode in Android 2.2.. 立. WiFi signal strength and get IP from the smart phone.. Another smart phone can scan. 學. ‧ 國. hotspot mode and it can transmit signal strength in figure 16.. Therefore, we can open TCP socket. between two smart phones and deliver the coordinate of them.. ‧. n. er. io. sit. y. Nat. al. Ch. engchi. Figure 16: Two HTC Desire smart phones.. i n U. (right opens hotspot mode). Figure 17 is an example for cooperative localization. location by nearby node N1 and N2.. N0 want to estimate or correct. N1 and N2 can scan the relative WiFi signal strength of. N0 and their possibility is higher than N0. distance D1 and D2 by equation 5.. v. Then we translate N1’s RSSIN0 and N2’s RSSIN1 to. Equation 5 is proposed by RADAR [5]. 22.

(31) d. P(d) = P(d0 ) − 10 ∗ n ∗ log (d ) + C. (5). 0. n indicates the rate at which the path loss increases with distance. at some reference distance.. P(d0) is the signal power. d is the transmitter-receiver separation distance.. And we calculate the possible location LN0(N1) within D1 from LN1 to LN0. possible location LN0(N2) within D2 from LN2 to LN0.. C is constant.. We also calculate. Finally we calculate weighted sum by. LN0, LN0(N1) and LN0(N2) and PN0, PN1 and PN2. So LN0c is the result location and PN0c is the. 政治大. highest possibility of PN0, PN1 and PN2. Figure 18 is the flowchart of cooperative localization.. 立. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 17: An example for cooperative localization.. 23.

(32) 立. 政治大. ‧. ‧ 國. 學 sit. y. Nat. io. n. al. er. Figure 18: Flow chart of cooperative localization.. Ch. engchi. 24. i n U. v.

(33) CHAPTER 4 Experimental Evaluation This chapter will be divided into four parts to be described.. 政治大. First, we make a description of. the setting of our simulation and creating a simulator in section 4.1.. 立. setting of implementation in section 4.3.. Third, we make a description of the. 學. ‧ 國. the simulation results are demonstrated specifically.. Second, in section 4.2,. Forth, we evaluate the accuracy of our algorithm in. implementation in section 4.4.. ‧ y. Nat. sit. 4.1 Simulation setup. n. al. er. io. To date many localization systems didn’t use simulator to test their performance, because signal strength fluctuations from minute to minute. simulator.. Ch. i n U. engchi. The simulator was implemented in Java and database is SQLite.. Figure 19 is the flow chart of our simulator.. First, we use HTC Desire(Android 2.2) to. site survey the 1/4 down-hill campus every 5~10 meter. survey.. v. But we measure the actual signal for our. Figure 20 is an example for site. Every point contains WiFi signal strength, longitude and latitude.. Second, we take. the HTC Desire and walk fifteen paths on campus and the Desire record actual longitude and latitude, WiFi signal strength and GPS longitude and latitude about 1 second.. 25.

(34) 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 19: Flow chart of simulator.. 26.

(35) 政治大. 立. ‧ 國. 學. Figure 20: An example of site survey.. Figure 21: An example of path.. ‧. There is a problem that how to simulate relative WiFi signal strength between users?. y. Nat. er. io. simulation.. sit. Therefore, third we measure WiFi signal strength every five meter using 2 HTC Desire before We save this to database and the simulator will match the distance to relative. n. al. Ch. signal strength between two paths.. i n U. v. Forth simulator creates and run fifteen path threads.. engchi. Fifth, simulator calculates distance between every pair of path by their actual longitude and latitude.. Sixth, simulator matches the distance to relative signal strength between them.. Figure 22, 23 and 24 are the distance to WiFi signal strength in different environments between two HTC Desire.. Red line (Meter to RSSI) is the actual WiFi signal strength. between two Desire and blue line(formula: RSSI to Meter) is equation 5.. We take every pair. of point on red line (Meter to RSSI) to equation 5 and calculate n=2.94. Therefore red line and blue line is closet match when n=2.94 in figure 22.. And n=1.93 and 3.02 in figure 23 and 24.. For example distance between two paths is 20 meter and simulator matches 20 meter to. 27.

(36) relative signal strength is -63.. And then cooperative localization calculates the distance by. -63. The equation 5 estimate the distance is 25 in figure 22. error is five meter.. In this example the estimation. In simulation we use figure 22 to in our simulator. And we use relative. WiFi signal strength when distance is 25 meter between two Desire, because we need to avoid there is a node that node’s localization possibility is very high and far away from user in cooperative localization.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 22: Distance (Meter) to WiFi signal strength (RSSI) in outdoor space. X axis is distance between two Desire and Y axis is relative WiFi signal strength.. 28.

(37) 立. 政治大. ‧. ‧ 國. 學. Nat. n. al. er. io. sit. y. Figure 23: Distance(Meter) to WiFi signal strength(RSSI) in indoor to outdoor space.. Ch. engchi. i n U. v. Figure 24: Distance(Meter) to WiFi signal strength(RSSI) in indoor space.. 29.

(38) Seventh simulator runs GPS, WiFi, heterogeneous and cooperative localization.. Simulator. load GPS longitude and latitude to simulate GPS localization and WiFi signal strength to simulate WiFi localization from path.. Finally, simulator calculates error distance between. actual location and the localization estimation of GPS, WiFi, Heterogeneous and Cooperative localization.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 25: WiFi localization possibility and error distance in all paths. Figure 25 is WiFi localization possibility and error distance in all paths.. This figure. shows that x axis is WiFi localization possibility and y axis is error distance of WiFi localization.. For example there is a point WiFi localization possibility is 0.63 and its error. distance is 9.. So this point possibility is high and error distance is short.. possibility is 0.35.. We set GPS. Because there are a little point error distance is bigger than 20 meter. when possibility is higher than 0.35.. When use GPS and WiFi localization simultaneously.. Heterogeneous and cooperative localization select GPS when WiFi localization possibility is 30.

(39) lower than 0.35.. 4.2 Simulation result We measure fifteen paths for simulation. section.. We will show four paths localization results in this. Figure 26 is the four paths on down-hill campus of NCCU.. percentage of error distance graph to show simulation result.. X axis is localization error. distance and Y axis is cumulative percentage of error distance (%).. 立. 政治大. ‧ 國. sit. io. PATH2 al v i n Ch engchi U. n. er. Nat. y. ‧. PATH1. PATH3. 學. PATH4. Figure 26: Four paths in simulation.. 31. We use cumulative.

(40) Figure 27 is simulation result for outdoor path (path 1). WiFi localization.. And heterogeneous localization can select GPS.. heterogeneous localization and GPS have similar accuracy. nearby path1.. GPS is more accurate than Therefore,. Furthermore there are two paths. They can correct the localization error of path1 by cooperative localization.. Cooperative localization has higher percentage than WiFi and heterogeneous localization at 15 meter to 20 meter.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 27: Simulation result in path 1. 32.

(41) Figure 28 is simulation result for path from indoor to outdoor to indoor (path 2). This path passes indoor space so GPS perform poorly in part of path.. Heterogeneous localization. used GPS in outdoor environments and used WiFi in indoor environments.. So. heterogeneous localization is more accurate than WiFi localization. Furthermore cooperative localization corrects the localization by nearby paths.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 28: Simulation result in path 2. 33.

(42) Figure 29 is a simulation result for indoor path (path 3). GPS performs poorly in indoor environments.. WiFi localization is very accurate in this building.. Heterogeneous. localization use human movement capacity and localization possibility to correct localization of WiFi localization. And other nearby path can’t help this path, because WiFi localization possibility is higher than other path in this path.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 29: Simulation result in path 3. 34.

(43) Figure 30 is the simulation result in indoor to outdoor path (path 4). There are more reliable neighbors help this path to correct localization error.. Cooperative localization. always has higher percentage than WiFi and heterogeneous localization.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 30: Simulation result in path 4. This section explains accuracy of cooperative localization and heterogeneous localization in outdoor path and indoor path and both of them.. Heterogeneous localization use. localization possibility to select more accurate location in different environments. its accuracy is higher than only WiFi localization or only GPS.. Therefore,. On the other hand, we use. localization possibility to decide which user is more accurate and they help us to correct localization.. So cooperative localization is always performs well in simulation result line. chart when there are reliable nearby paths. 35.

(44) 4.3 Implementation setup We want to demonstrate our methods can be used on smart phone and they are more accurate in large scale environments. localization in Android.. So we implemented WiFi, heterogeneous and cooperative. In implementation we separate indoor and outdoor experimental.. User’s actual location is GPS longitude and latitude in outdoor space but was clicked on google map by user in indoor space. these actual locations.. We calculate error distance between localization and. And the site survey database is the same as simulation.. a picture of our experimental.. 立. 政治大. We walk three paths on campus.. path 1 and this path is in outdoor space.. Second path is the same as simulation. Third path is in our department building and this. User walks these paths and two other people are always nearby him. ‧. and share their locations.. In implementation we simulate IOT localization.. Nat. And then write the longitude and latitude to button.. sit. longitude and latitude on every building.. First, we get. y. path is in indoor space.. First path is 1/4 down-hill. 學. ‧ 國. campus of NCCU and this path is in outdoor space.. Figure 31 is. al. n. building.. er. io. When user passes the building and clicks the button to get the longitude and latitude on that. Ch. engchi. 36. i n U. v.

(45) 立. 政治大. ‧ 國. 學 ‧. Figure 31: A picture of our experimental.. io. sit. y. Nat. (Left picture is cooperative localization result on google map, middle picture is nearby people and test cooperative localization user, right is console mode of our program). n. al. er. 4.4 Implementation result. Ch. i n U. v. We also use cumulative percentage of error distance graph to show implementation result.. engchi. X axis is localization error distance and Y axis is cumulative percentage of error distance (%).. 37.

(46) Figure 32 show the implementation result in outdoor path. 14% points’ error distance are zero meters.. We use IOT localization so. In this case we open GPS every 10~20 second.. Therefore, heterogeneous localization can use some GPS locations to get beyond accuracy of WiFi localization. Furthermore there always two people nearby user and they deliver their GPS location to the user.. This causes that cooperative localization is more accurate than. other localization algorithms.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 32: First implementation result in outdoor path.. 38.

(47) Figure 33 is an implementation result of outdoor path the same as path1 in simulation. In this path we open GPS every 5~8 second. We discover accuracy of cooperative and heterogeneous localizations are similar in this path. And we add one line that is heterogeneous localization but it disables GPS. Because we want to verify GPS is very accurate in this path and other nearby people can’t help user to achieve more accurate.. The. accuracy of heterogeneous localization is to get beyond to a lot than disable GPS one. On the other hand, WiFi localization possibility is always low in this path. So heterogeneous. 政治大. localization disable GPS and WiFi localization’s possibility are almost.. 立. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 33: Second implementation result in outdoor path.. 39.

(48) Figure 34 is an implementation result of indoor path in DaRen building. There are about 18 access points in DaRen Building. Therefore accuracy is very high of WiFi localization. And heterogeneous localization always selected WiFi localization. But cooperative localization doesn’t use other nearby people localization to correct location, because localization possibility of WiFi localization is higher than nearby people in this In this case we use Figure 23 to translate relative signal strength to distance.. And n is 1.93 in equation 5.. 立. 政治大. ‧. ‧ 國. 學. io. sit. y. Nat. n. al. er. environment.. Ch. engchi. i n U. v. Figure 34: Implementation result in indoor path.. 40.

(49) In this section we implement heterogeneous and cooperative localization on HTC Desire. We take them walk on campus actually.. The experimental results show that heterogeneous. localization in outdoor environments is better than WiFi localization.. Furthermore. cooperative localization is more accurate than other localizations if there are reliable nearby neighbors.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. 41. i n U. v.

(50) CHAPTER5 Conclusions In this paper, we proposed heterogeneous and cooperative localizations. We implement a. 政治大. WiFi localization system with probabilistic approach in indoor environments and the accuracy. 立. is about 4 meter in Nangang Exhibition Hall.. Then we use localization possibility to. ‧ 國. 學. combine GPS, WiFi and IOT localization to proposed heterogeneous localization in large scale environments.. ‧. io. They share their locations to correct user’s localization.. al. n. localization.. sit. If their localization results are more reliable. Therefore, we proposed cooperative. er. Nat. scan other user’s relative WiFi signal strength.. y. Furthermore, we consider if user can’t use GPS, WiFi and IOT localization and they can. i n U. v. We use localization possibility to decide who is more reliable.. deliver their location to user.. Ch. engchi. Then they can. And user use relative signal strength and the locations to. estimate a new location. Finally, we evaluate the accurate of our algorithm in simulation and implementation. simulation, we measure actual signal data in down-hill campus of NCCU and run our algorithm using the data.. Then heterogeneous and cooperative localizations are more. accurate than other localization method.. In implementation, we verify that our idea can be. implemented on smart phones and our algorithms are more accurate.. All the results show. that heterogeneous and cooperative localizations is more applicable than WiFi localization and GPS in large scale environments. 42. In.

(51) Reference [1] J. Hightower, R. Want, and G. Borriello, ―SpotON: An indoor 3D location sensing technology based on RF signal strength,‖ Univ. Washington, Seattle, Tech. Rep. UW. 政治大 L. M. Ni,Y. Liu,Y. C. Lau, and A. P. Patil, ―LANDMARC: Indoor location sensing 立. CSE 2000–02-02, Feb. 2000. [2]. ‧ 國. 學. using active RFID,‖ Wireless Netw., vol. 10, no. 6, pp. 701–710, Nov. 2004. [3] Bahl, P., Padmanabhan, V.N.: ―Radar an in-building RF-based user location and tracking. ‧. system.‖ In: INFOCOM 2000, Tel Aviv, Israel, pp. 775–784 (2000). sit. y. Nat. [4] M. Youssef and A. K. Agrawala, ―Handling samples correlation in the Horus system,‖. io. er. IEEE INFOCOM 2004, Hong Kong, vol. 2, pp. 1023–1031, Mar. 2004.. al. v i n Haeberlen, A.; Flannery, E.;C Ladd, Rudys, A.; Wallach, D.;and Kavraki, L. ―Practical h eA.; ng chi U n. [5] Ekahau, Inc. Ekahau Positioning Engine 2.0. http://www.ekahau.com/ [6]. robust localization over largescale 802.11 wireless networks.‖ In Proc. of the Tenth ACM International Conference on Mobile Computing and Networking 2004. [7] H. Lemelson, S. Schnaufer and W. Effelsberg, ―Automatic Identification of Fingerprint Regions for Quick and Reliable Location Estimation.‖ Pervasive Computing and Communications Workshops (PERCOM Workshops) 2010. [8] Dik Lun Lee and Qiuxia Chen, ―A Model-Based WiFi Localization Method‖, The Hong Kong University of Science and Technology, INFOSCALE 2007 June 6-8, 2007, Suzhou, China, ACM 2007. 43.

(52) [9] R. Hansen and B Thomsen, ―Efficient and Accurate WLAN Positioning with Weighted Graphs‖, MOBILIGHT 2009, LNICST 13, pp. 372–386, 2009. [10] O. Baala, Y. Zheng, A. Caminada. "The Impact of AP Placement in WLAN-Based Indoor Positioning System," Proceedings of the 8th International Conference on Networks, Cancun, Mexico, pp.12-17, 2009. [11] P. Bolliger, K. Partridge, M. Chu, and M. Langheinrich. ―Improving Location Fingerprinting through Motion Detection and Asynchronous Interval Labeling.‖ In Proc.. 政治大. Location and Context Awareness, pp. 37–51, Tokyo, Japan, May 2009.. 立. [12] Bolliger, P.: Redpin – ―adaptive, zero-configuration indoor localization through user. ‧ 國. 學. collaboration.‖ In: Workshop on Mobile Entity Localization and Tracking in GPS less Environment Computing and Communication Systems (MELT), San Francisco(2008). ‧. [13] V. Honkavirta, T. Perala, S. Ali-Loytty, and R. Piche. ―A comparative survey of wlan. Nat. n. al. er. io. Communication WPNC 2009, pp. 243–251, 2009.. sit. y. location fingerprinting methods.‖ In Proc. 6th Workshop on Positioning, Navigation and. i n U. v. [14] Hendrik Lemelson, Stephan Kopf, Thomas King, Wolfgang Effelsberg, "Improvements. Ch. engchi. for 802.11-Based Location Fingerprinting Systems," compsac, vol. 1, pp.21-28, 2009 33rd Annual IEEE International Computer Software and Applications Conference, 2009 [15] Liu, H., H. Darabi, P. Banerjee, and J. Liu, "Survey of wireless indoor positioning techniques and systems," IEEE Transactions on systems, Man, and Cybernetics — Part C: Applications and Reviews, Vol. 37, No. 6, November 2007. [16] Chan L-W, Chiang J-R, Chen Y-C, Ke C-N, Hsu J, Chu H-H (2006) ―Collaborative localization—enhancing WiFi-based position estimation with neighborhood links in clusters.‖, in Proceedings of the International conference on Pervasive Computing (PERVASIVE 2006), pp. 50–66 44.

(53) [17] N. Banerjee, S. Agarwal, P. Bahl, R. Chandra, A. Wolman, and M. Corner. ―Virtual compass: relative positioning to sense mobile social interactions.‖ In Pervasive, 2010. [18] ORBI positioning system. http://ssrc.nccu.edu.tw/orbi/ [19] News clip for ORBI. http://mag.udn.com/mag/digital/storypage.jsp?f_MAIN_ID=320&f_SUB_ID=4996&f_ ART_ID=321825 [20] H. Lemelson, M. B. Kjæ rgaard, R. Hansen, and T. King. ―Error Estimation for Indoor. 政治大. 802.11 Location Fingerprinting.‖ In Proc. Location and Context Awareness, pp. 138–155,. 立. Tokyo, Japan, May 2009.. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. 45. i n U. v.

(54)