臺灣博碩士論文加值系統

由於近年來因國際情勢的緊張，許多國家彼此間衝突不斷，舉凡俄烏戰爭、以哈戰爭皆使用到無人機進行偵察與打擊敵方目標的作業，因此在無人機上皆會搭載鏡頭與目標物辨識功能，倘若能結合機間通訊功能，當偵察機辨視出目標物並對其進行定位，透過無線通訊功能回傳座標給地面站，再請求火力支援去打擊目標，以發揮無人機在不對稱作戰中，制敵的關鍵利器。
為了能達成上述想定，每台無人機搭載嵌入式系統，透過ROS系統進行機間通訊，並使用MAVROS函式庫來與飛控連接。採用此架構能在機載電腦上進行即時運算與控制，同時透過UDP通訊協議，將無人機飛控資料回傳至地面站做監控。
長機負責接收垂直向下式偵察機回傳的目標座標，並且飛往該目標物上空進行投擲。僚機會與長機通訊，透過ROS(Robot Operation System)系統下的機間通訊功能，長機將進行相對位置的計算，並將計算完成後的位置發送給僚機，使僚機能與長機保持距離進行編隊的飛行，長機接收到目標物位置資訊時僚機會同時起飛並且依照需求進行編隊飛行，當長機與僚機到達目標物上空後，地面站需按下確認按鈕後無人機會同時進行投擲的動作，投擲完成後則會回到起飛位置等待偵察機傳送新一筆目標物位置。並設計一套地面站負責監視所有無人機的狀態、目標物座標與圖片、機隊是否到達目標物位置，並點選按鈕使機隊能同時投擲。
最終，本研究達成了全自動飛行目的，垂直向下式偵察機進行自動偵蒐目標並將目標位置座標與圖片回傳至地面站，長機接收到座標訊息後，與二台僚機同時起飛，僚機會依據長機位置進行相對位置的編隊，到達目標物上空時需切換成投擲隊形地面站按鈕才能點選，點選按鈕後長機與僚機可以同時投擲。在地面站畫面內可監視各台無人機的基本數據、目標物圖片與座標、投擲按鈕介面設計與地圖顯示無人機位置與目標物位置，並提升了垂直向下式偵蒐目標物之精準度，在高度40公尺、50公尺、60公尺下與目標物之距離誤差分別為0.15公尺、0.48公尺、0.54公尺；斜向式偵蒐目標物之精準度為2.1公尺、1.8公尺、1.66公尺
並進行了斜向偵測式與垂直向下式偵察機的誤差比較與提升精準度的探討。

Due to the recent tensions in international situations, conflicts between many countries have been ongoing. In the Russia-Ukraine war and the Israel-Hamas conflict, drones have been used for reconnaissance and striking enemy targets. These drones are equipped with cameras and target recognition functions. If communication between drones can be integrated, a reconnaissance drone could identify a target and pinpoint its location, then transmit the coordinates to the ground station via wireless communication. The ground station could then request fire support to strike the target, making drones a crucial weapon in asymmetric warfare.
To achieve the above scenario, each drone is equipped with an embedded system that communicates through the ROS (Robot Operating System) and uses the MAVROS library to connect with the flight control system. This architecture allows real-time computation and control on the onboard computer, while using the UDP communication protocol to send drone flight control data back to the ground station for monitoring.
The lead drone is responsible for receiving the target coordinates from the vertically oriented reconnaissance drone and flying over the target for a strike. The wingmen communicate with the lead drone using ROS's inter-drone communication function. The lead drone calculates the relative position and sends it to the wingmen, allowing them to maintain formation flight with the lead drone. When the lead drone receives the target position information, the wingmen take off simultaneously and fly in formation according to the mission requirements. Once they reach the target, the ground station needs to press a confirmation button to enable the drones to release their payloads simultaneously. After releasing, they return to their launch position to await new target coordinates from the reconnaissance drone.
A ground station is designed to monitor all drones' statuses, target coordinates, images, and whether the fleet has reached the target. By pressing a button, the fleet can release their payloads simultaneously. Ultimately, the study achieved fully autonomous flight. The vertically oriented reconnaissance drone automatically searches for targets and sends target coordinates and images back to the ground station. Upon receiving the coordinates, the lead drone and two wingmen take off simultaneously. The wingmen form a relative position with the lead drone. When they reach the target, they switch to the drop formation, and the ground station button can be pressed. After pressing the button, the lead and wingmen can release their payloads simultaneously. On the ground station interface, users can monitor each drone's basic data, target images, coordinates, release button design, and map display of drone and target positions. The accuracy of vertical reconnaissance of targets improved, with errors of 0.15 meters, 0.48 meters, and 0.54 meters at altitudes of 40 meters, 50 meters, and 60 meters, respectively. The accuracy for oblique reconnaissance of targets was 2.1 meters, 1.8 meters, and 1.66 meters. A comparison of errors between oblique detection and vertical reconnaissance drones was conducted, along with a discussion on improving accuracy.

摘要...........................................................................i
Abstract.....................................................................iii
誌謝...........................................................................v
目錄..........................................................................vi
表目錄.........................................................................x
圖目錄........................................................................xi
符號說明 ....................................................................xvii
第一章 緒論................................................................- 1 -
1.1前言 ..................................................................- 1 -
1.2研究動機................................................................- 2 -
1.3論文架構................................................................- 2 -
第二章 文獻回顧.............................................................- 3 -
2.1 邊緣運算與嵌入式系統................................................- 3 -
2.2 AI影像辨識.........................................................- 4 -
2.3 無人機影像追蹤......................................................- 6 -
2.4 雷射測距與斜向偵測..................................................- 6 -
2.5 無人機群飛之機間通訊與網路...........................................- 9 -
2.6 多旋翼無人機編隊...................................................- 12 -
第三章 研究硬體應用及架構介紹...............................................- 16 -
3.1無人機硬體配置..........................................................- 16 -
3.2Pixhawk Orange Cube與Pixhawk 6X飛行控制電腦.............................- 19 -
3.3RTKGPS.................................................................- 21 -
3.4HereLink數圖傳一體遙控器................................................- 22 -
3.5AT9S遙控器.............................................................- 23 -
3.6嵌入式電腦.............................................................- 24 -
3.6.1Nvidia Jetson Xavier NX.............................................- 24 -
3.6.2Nvidia Orin Nano....................................................- 24 -
3.6.3Nvidia Jetson Nano..................................................- 25 -
3.6.4嵌入式電腦性能比較....................................................- 25 -
3.7 相機雲台..............................................................- 26 -
3.7.1 AXG-AS03F單光三軸變焦相機............................................- 26 -
3.7.2 KTG-TT30三光三軸變焦相機.............................................- 27 -
第四章研究軟體應用與架構介紹................................................- 29 -
4.1Mission Planner地面監控站..............................................- 29 -
4.2 Ardupilot .................................................- 31 -
4.3 MAVLINK通訊協定 .................................................- 31 -
4.4 ROS..............................................................- 32 -
4.5 MAVROS...........................................................- 33 -
4.6 GAZEBO模擬器介紹 .................................................- 34 -
4.7 影像辨識與模型訓練.................................................- 35 -
4.7.1 Yolov4-tiny 運算架構..............................................- 35 -
4.7.2 LabelIMG 標註軟體.................................................- 36 -
4.7.3 Darknet_ROS介紹...................................................- 37 -
4.8 Tkinter GUI雲台控制介面設計........................................- 38 -
4.9 Node-Red.............................................................- 39 -
4.10基於ROS下的機間通訊介紹................................................- 39 -
第五章AI影像辨識與目標定位任務規劃 .........................................- 44 -
5.1 垂直式向下偵察機之偵察方式.........................................- 44 -
5.1.1最小中心框追蹤方式....................................................- 44 -
5.1.2影像追蹤流程圖........................................................- 45 -
5.1.3 垂直向下式影像修正方式與精準度提高之方法研究..........................- 46 -
5.1.4垂直向下式影像座標計算目標經緯度法......................................- 50 -
5.1.5影像中計算目標物座標方法...............................................- 50 -
5.2 斜向式偵察機之偵察方式..............................................- 54 -
5.2.1中心點之追蹤方式......................................................- 54 -
5.2.2雲台追蹤流程圖........................................................- 55 -
5.2.3 斜向偵測式之精準度提高之方法研究....................................- 56 -
第六章機間通訊與群組飛行....................................................- 58 -
6.1 機間通訊設定...........................................................- 58 -
6.2機間通訊測試............................................................- 60 -
6.3 編隊隊形計算...........................................................- 61 -
6.4群組飛行流程規劃........................................................- 63 -
6.4.1長機流程規劃..........................................................- 63 -
6.4.2僚機流程規劃..........................................................- 65 -
6.5群組編隊飛行模擬........................................................- 66 -
6.6群組編隊實際飛行........................................................- 68 -
第七章 實驗結果與分析.....................................................- 69 -
7.1 影像圖資訓練與測試....................................................- 69 -
7.2 垂直向下偵測與斜向偵測之結果........................................- 72 -
7.3 群組無人機之飛行測試結果............................................- 76 -
7.4 地面站軟體介面.....................................................- 77 -
7.5 最終整合測試流程...................................................- 79 -
7.6 實驗結果分析.......................................................- 82 -
7.6.1 垂直向下式單目標影像辨識定位精準度.....................................- 82 -
7.6.2 垂直向下式單目標影像辨識，定位使用RTK誤差測量分析.......................- 86 -
7.6.3 垂直向下式多目標影像辨識定位精準度.....................................- 92 -
7.6.4 斜向偵測式影像辨識定位精準度 .........................................- 102 -
7.6.5斜向偵測式影像辨識定位使用RTK誤差測量分析..............................- 105 -
7.6.6 斜向偵測式多目標影像辨識定位誤差測量分析..............................- 107 -
7.6.7 垂直向下式定位與斜向式定位精準度比較..................................- 116 -
7.7. 各嵌入式電腦使用YoloV4-tiny之執行效率比較.............................- 119 -
7.8 雲台自動掃描與多目標追蹤定位功能實踐....................................- 121 -
第八章 結論與展望.........................................................- 128 -
參考文獻 .................................................................- 132 -
Extended Abstract........................................................- 134 -
Chapter 1 Introduction...................................................- 136 -
Chapter 2 Research Method................................................- 137 -
2.1 Study Design.....................................................- 137 -
2.2 Materials and Equipment:.........................................- 138 -
Chapter 3. Swarm System Architecture.....................................- 140 -
3.1 Swarm Architecture...................................................- 140 -
3.2 Leader drone Flowchart...............................................- 141 -
3.3 Follower Flowchart...................................................- 142 -
3.4 Follow-up Method.................................................- 143 -
Chapter 4 Vertical Reconnaissance System Architecture....................- 146 -
4.1 Vertical Reconnaissance Drone Flowchart..............................- 146 -
4.2Principle for Calculating Target Coordinates in an Image..............- 147 -
Chapter 5 The Oblique Detection System Architecture......................- 150 -
5.1 Oblique Reconnaissance Drone Flowchart...........................- 150 -
5.2 Principle of Oblique Detection.......................................- 151 -
Chapter 6 Experiment results.............................................- 153 -
6.1 Swarm drone .........................................................- 153 -
6.2 Vertical Reconnaissance Positioning Accuracy.....................- 154 -
6.3 Oblique Reconnaissance Positioning Accuracy......................- 155 -
Chapter 7 Conclusion.....................................................- 157 -
References...............................................................- 158 -

[1] https://blogs.nvidia.com/blog/metropolis-jetson-isaac-robotics-edge-ai-developers/?ncid=em-news-815061-vt48 NVIDIA 擴展機器人平台以因應生成式 AI 的興起。
[2] 蘇孟宗，陳右怡，“人工智慧驅策臺灣產業跨域創新”， 國土及公共治理季刊 ； 6卷4期， P40 – 49， 2018 / 12 / 01。
[3] 洪晨鈞「基於 YOLO v4-tiny 演算法對抽菸進行影像偵測」。碩士論文，國立臺北科技大學電子工程系，2022。
[4] 徐光廷。「人工智慧與邊緣運算技術應用於無人機之影像追蹤」。碩士論文，國立虎尾科技大學飛機工程系航空與電子科技碩士班，2022。
[5] 邱志成。「具UAV追瞄與座標定位之旋轉載台的反無人機系統研究」。博士論文，國立高雄科技大學工學院工程科技博士班，2018。
[6] Chen, Xi, Jun Tang, and Songyang Lao. 2020. "Review of Unmanned Aerial Vehicle Swarm Communication Architectures and Routing Protocols" Applied Sciences 10, no. 10: 3661.。
[7]平行機群概念框架與應用 沈震,劉雅婷,董西松,白天翔,胡斌,熊剛,王飛躍
中國科學院自動化研究所複雜系統管理與國家重點實驗室,青島智能產業技術研究學院,北京市智能化技術與系統工程技術研究中心,中國科學院大學,國防科技大學軍事計算與平行系統技術研究中心。
[8]陳冠銘。「具AI邊緣運算與機間通訊功能之蜂群無人機系統之研製」。碩士論文，國立虎尾科技大學飛機工程系航空與電子科技碩士班，2023。
[9]Ardupilot. (n.d.). Cube Orange飛控介紹. Retrieved from
https://ardupilot.org/copter/docs/common-thecubeorange-overview.html
[10]Ardupilot. (n.d.). Pixhawk6x飛控介紹. Retrieved from
https://ardupilot.org/plane/docs/common-holybro-pixhawk6X.html
[11]Drones Boy. (n.d.). RTK工作原理介紹. 取自
https://dronesboy.com/ppk-or-rtk-which-is-best-in-drone-mapping/
[12] CubePilot. (n.d.). HereLink介紹. Retrieved from
https://docs.cubepilot.org/user-guides/herelink/herelink-overview
[13] Radiolink. (n.d.). AT9S遙控器介紹.
https://www.radiolink.com/at9spro
[14]T客邦. (n.d.). Orin Nano性能規格介紹. 取自
https://www.techbang.com/posts/105172-jetson-orin-nano-development-kit
[15] NVIDIA. (n.d.). Jetson Nano介紹. 取自
https://www.nvidia.com/zh-tw/autonomous-machines/embedded-systems/jetson-nano/product-development/
[16]Ardupilot. (n.d.). Retrieved from https://ardupilot.org/
[17]PX4. (n.d.). Retrieved from https://docs.px4.io/main/zh/
[18]知乎. (2021, September 29). MAVLink介紹.取自https://zhuanlan.zhihu.com/p/415266156
[19] CSDN. (2016, October 12). ROS介紹. 取自 https://blog.csdn.net/aitazhixin/article/details/52794567
[20] KeywordSEO. (n.d.). YoloV4-tiny介紹. 取自
https://www.keywordseo.com.tw/blog1/yolov4-vs-yolov4-tiny/
[21] GT Wang. (2017, November 29). LabelImg標註軟體介紹.取自https://blog.gtwang.org/useful-tools/labelimg-graphical-image-annotation-tool-tutorial/
[22] GitHub. (n.d.). 本研究所使用之Darknet_ROS功能包. Retrieved from
https://github.com/Tossy0423/yolov4-for-darknet_ros?tab=readme-ov-file
[23] CSDN. (2021年1月11日). ROS通訊原理介紹. 取自
https://blog.csdn.net/weixin_45661757/article/details/111573648
[24] Edu For Everyone. (n.d.). 海拔高與相對高.取自
https://eduforeveryone123.wixsite.com/flippinggeography/blog?fbclid=IwAR2dQe910Xc9RUtpxM6P0tdgh6A-qfxoT-poxj-IlraddI08TWpdtBuWnGY
[25] 黃博信. (2022). 淺談新式測距經緯儀於砲兵基地測地作業之運用. 《陸軍砲兵季刊》, 199, 17-21.