近年來科技的進步，人手一支智慧型手機已習以為常，也因智慧型手機的普及，相關的應用程式也數以萬計的增長。在手機提供的應用程式服務中，許多程式都需要使用到定位資訊所提供的資料，因此如何得到準確的定位資料就非常重要。而GPS 是目前室外應用中最為廣泛的定位系統，但在室內定位時會因為建築物而降低訊號強度，使得室內定位變得不準確，因此，有許多得室內定位方法被提出。
目前室內定位的方法有很多種，例如:WiFi定位、接收信號角度定位法(AOA)、到達時間差(TDOA)、接收信號強度指標(RSSI)、訊號紋比對（模式匹配）、iBeacon 定位系統、藍芽訊號定位等等的室內定位方法，但是WiFi、 Angle of Arrival, (AOA)等等的定位方法精準度不高，而iBeacon、Radio Frequency Identification(RFID)等等的定位方法需要布置大量的iBeacon及RFID設備，需要的成本也比較高，因此本文將提出一種複合式室內定位方法來提升定位的精準度及降低定位成本。
我們透過WiFi RSSI的資料收集，以及深度學習來判斷其室內位置，我們在實驗環境中架設3台WiFi AP，每隔一公尺設一個參考點收集RSSI數據，將數據進行整理後輸入至MLP(Multilayer Perceptron)進行訓練及學習，經過MLP(Multilayer Perceptron)學習後準確率約為6成。有了MLP的定位資料後，我們用航位推算法加上深度學習，依照測試目標的上個位置的資訊與陀螺儀的資訊推斷出測試目標下一個位置可能出現的範圍，並將出現的範圍套用在MLP所推算出的位置，將範圍外的數值去除，以達到更準確的定位路徑。
本論文研究結果顯示，如果單獨使用WiFi定位的話，誤差的距離約為兩公尺，但是用本文所提出的方法能將誤差的距離降低為一點五公尺，且使用航位推算加上深度學習的方式，可以去除MLP所推算出的錯誤定位資料改善室內定位的誤差。

In recent years, the advancement of technology, people are accustomedto use smart phone everyday, and because of the popularity of smart phones, related applications have also grown tens of thousands. In the application services provided by mobile phones, many programs need to use the information provided by the positioning information, so how to get accurate positioning information is very important, and GPS is the most widely used positioning system in outdoor applications. However,the signal strength will be reduced due to the building, and the indoor positioning will become inaccurate. Hence, many indoor positioning methods are proposed to improve the accuracy of indoor positioning.
At present, there are many methods for indoor positioning, such as: WiFi positioning, received signal Angle of Arrival positioning (AOA), Arrival Time difference (TDOA), received signal strength indicator (RSSI), signal pattern matching, iBeacon positioning system, Bluetooth, etc..But some of those methods, such as: WiFi, Angle of Arrival, (AOA), cannot providehigherpositioning accuracy; and some of them, such as:iBeacon, Radio Frequency Identification (RFID), need to arrange a large number of iBeacon and RFID devices to achieve higher accuracy and results in higher cost. Therefore, this paper proposes a composite indoor positioning method. to improve positioning accuracy and reduce positioning costs.
We collect WiFi RSSI data and feed data into deeplearning method to determinetarget’sindoor location. We set up a total of 3 WiFi APs in experimental environment, and set up a reference point per meterapart to collect RSSI data. Then, after the data preprocessing, feed the organized data into multilayer perceptron method for training and learning. After using MLP (Multilayer Perceptron) learning method, the accuracy of positioning is about 60%.Hence, we use the dead reckoning algorithm plus MLP to infer the possible range of the next position of the test target according to the information of the last position of the test target, and apply the range that are estimated by the MLP. The estimated locations which areoutside the range will beremoved to achieve a more accurate positioning location and moving path.
As the experimental results show that if the WiFi positioning is used alone, the positioning error is about two meters, but the method proposed in this thesiscan reduce the error to around 1.5 meters.Moreover,use the dead reckoning plus MLPcan reduce the estimation error of MLP and improve the indoor positioning accuracy.

目錄
致謝	I
論文提要內容	Ⅱ
Abstract	Ⅳ
目錄	V
圖目錄	VII
表目錄	IX
第一章 緒論	1
1.1前言	1
1.2動機與目的	1
1.3 論文章節架構	2
第二章 相關研究及背景資料	3
2.1接收信號角度定位法（到達角度，Angle of Arrival，AOA）	3
2.2到達時間定位法（到達時間差，TDOA）	4
2.3訊號指紋比對（模式匹配）	4
2.4 iBeacon	5
2.5其他定位方法	6
第三章基於深度學習及行人路徑推算室內定位方法	9
3.1 RSSI與多層感知器定位法	9
3.1.1 深度學習—多層感知器	10
3.1.2 MLP 定位成果與限制	16
3.2行人航位推算法	18
3.2.1 陀螺儀數值收集環境	18
3.2.2陀螺儀數值分析	20
3.3 基於行人路徑推算及深度學習室內定位方法	24
第四章  實驗結果	29
4.1實驗內容	29
4.2實驗結果與分析	30
4.2.1實驗路徑一	31
4.2.2實驗路徑二	37
4.2.3實驗路徑三	42
4.2.4實驗路徑四	46
4.2.5實驗路徑五	52
第五章  貢獻與未來展望	57
5.1 主要貢獻	57
5.2未來展望	57
參考文獻	58

圖2.1 Angle of Arrival，AOA	3
圖2.2 TDOA	4
圖2.3iBeacon設備	6
圖2.4可見光定位示意圖	8
圖3.1多層感知器	10
圖3.2接收信號強度指標，RSSI	11
圖3.3實驗環境	13
圖3.4 RSSI值訊號	14
圖3.5將實驗環境分成十九個測試點	15
圖3.6 RSSI收集及輸入至MLP流程圖	16
圖3.7 MLP模型訓練後所猜測的數值(第1至500個)	17
圖3.8 Arduino UNO板	19
圖3.9實驗路徑	20
圖3.10陀螺儀X軸數值	21
圖3.11陀螺儀Y軸數值	22
圖3.12陀螺儀推算位置流程圖	23
圖3.13基於行人路徑推算及深度學習室內定位方法流程圖	24
圖3.14實驗路徑圖	25
圖3.15陀螺儀數值圖	26
圖3.16MLP模型推算數值201~301	27
圖3.17MLP模型推算數值501~601	28
圖4.1實驗環境座標圖	30
圖4.2實驗路徑一座標圖	31
圖4.3陀螺儀數值	33
圖4.4 MLP模型推算準確率為0.54	34
圖4.5實驗路徑一MLP測試結果	34
圖4.6 MLP模型計算結果	35
圖4.7實驗路徑一的三種方法測試結果	36
圖4.8實驗路徑二座標圖	37
圖4.9陀螺儀數值圖	38
圖4.10 MLP模型推算的準確率為0.548	39
圖4.11實驗路徑二MLP測試結果	39
圖4.12 MLP模型計算結果	40
圖4.13實驗路徑二的三種方法測試結果	41
圖4.14實驗路徑三座標圖	42
圖4.15陀螺儀數值圖	43
圖4.16 MLP模型推算的準確率為0.66	44
圖4.17實驗路徑三MLP測試結果	44
圖4.18 MLP模型計算結果	45
圖4.19實驗路徑三的三種方法測試結果	46
圖4.20實驗路徑四座標圖	47
圖4.21陀螺儀數值	48
圖4.22 MLP模型推算的準確度為0.670	49
圖4.23實驗路徑四MLP測試結果	49
圖4.24 MLP模型計算結果	50
圖4.25實驗路徑四的三種方法測試結果	51
圖4.26實驗路徑五座標圖	52
圖4.27實驗路徑五RSSI測試點	53
圖4.28實驗路徑五的兩種方法測試結果	54
圖4.29實驗路徑五的兩種方法測試結果	55

表2.1 室內定位方法比較	8

[1]接收信號角度定位法，卓尚澤，http://designer.mech.yzu.edu.tw/articlesystem/article/compressedfile/(2009-07-03)%20%E5%AE%A4%E5%85%A7%E5%AE%9A%E4%BD%8D%E6%8A%80%E8%A1%93%E7%B0%A1%E4%BB%8B.aspx?ArchID=961
[2] Pratap N. Misra, “The Role of the Clock in a GPS Receiver”, GPS World, April 1996.
[3]訊號指紋比對（模式匹配），https://kknews.cc/other/ekar4mn.html
[4]Mohd Ezanee Rusli ; Mohammad Ali ; Norziana Jamil ; Marina Md Din， An Improved Indoor Positioning Algorithm Based on RSSI-Trilateration Technique for Internet of Things (IOT)， 2016 International Conference on Computer and Communication Engineering (ICCCE)
[5] Guodong Yang ; Jikai Zhang， Research on Indoor Positioning Accuracy Based on Wi-Fi Intensity Values， 2016 International Conference on Network and Information Systems for Computers (ICNISC)
[6] Andy Cavallini, “iBeacons Bible 1.0”, Retrieved from:
https://meetingofideas.files.wordpress.com/2013/12/ibeacons-bible-1-0.pdf
[7] Linjun Yu ; Yalan Liu ; Tianhe Chi ; Lin Peng， An iBeacon-based indoor and outdoor positioning system for the fire emergency command， 2017 Forum on Cooperative Positioning and Service
[8] Zhiqiang He ; Binyue Cui ; Wei Zhou ; Shigeki Yokoi， A proposal of interaction system between visitor and collection in museum hall by iBeacon， 2015 10th International Conference on Computer Science & Education
[9] Sruthi Menon ; Anu George ; Nehal Mathew ; Vishnu Vivek ; Jisha John， Smart workplace — Using iBeacon， 2017 International Conference on Networks & Advances in Computational Technologies (NetACT)
[10 ] Zigbee，https://www.zigbee.org/zigbee-for-developers/zigbee-3-0/
[11] Keigo Nomura ; Fumiaki Sato， A Performance Study of ZigBee Network under Wi-Fi Interference， 2014 17th International Conference on Network-Based Information Systems
[12] Hisanori Matsumoto ; Haruka Kusano ; Tatsushi Morokuma ; Ken Sakamura， Numerical and experimental investigation of TDOA-based positioning system by ultra-wideband impulse radio， 2011 IEEE Topical Conference on Wireless Sensors and Sensor Networks
[13] 可見光通訊技術，劉彥良/盧奕誠，https://zh.wikipedia.org/wiki/%E5%8F%AF%E8%A6%8B%E5%85%89%E9%80%9A%E8%A8%8A
[14] Youngsuk Kim ; Junho Hwang ; Jisoo Lee ; Myungsik Yoo， Position estimation algorithm based on tracking of received light intensity for indoor visible light communication systems， 2011 Third International Conference on Ubiquitous and Future Networks (ICUFN)
[ 15] MLB's iBeacon Technology is A Marketing Home Run. Retrieved from: https://www.ama.org/publications/MarketingNews/Pages/mlb-ibeacons.aspx
[16 ] 行人航位推算系統，張永超，https://robinchao.github.io/2017/11/25/pdr-indoor-position.html
[17] Arduino UNO. Retrieved from: https://www.arduino.cc/
[18] MPU-6050，https://www.invensense.com/products/motion-tracking/6-axis/mpu-6050/