現今工業發展日新月異，隨著人力成本地提高，工業4.0成為工業界的熱門話題，建立起從生產、銷售到售後保固都能一條龍兼顧的雲端數據庫，以減少人力的支出、提高生產效率，成為各個公司追求的目標。
工業4.0帶來的改變遠遠不只如此，還有無人超市、無人駕駛…等「無人化」的出現，機器人取代勞力逐漸成為趨勢，而最重要的莫過於無人工廠，偌大的工廠只有寥寥數十人監控機器人的工作狀態，甚至可以做到24小時關燈、不停機生產，極大地節省勞力成本，本篇論文就在探討如何利用ROS建立一個具有機器視覺與物件處理的移動機器人系統，包含了室內的避障、路徑規劃，而三維機器視覺能進行物件辨識並配合機械手臂的取放功能，使可以實際應用於無人工廠的移載應用機器人。
使用三維點雲當作機器視覺的輸入端，優點在於能得到比二維影像更精準的物體姿態估算，搭配不同的三維機器視覺演算法，讓實驗結果比使用單一的演算法來的精準和迅速。最後，會將上述結合的演算法搭配機械手臂做真實物體的位姿估算及取放分類，驗證其實用性。

Industrial development is ever-changing. With the increase in the cost of labor, Industry 4.0 has become a popular issue in the industry. It creates a cloud database that provides a streamlined business that includes manufacturing, marketing, warranty, and after-sales services. To reduce the cost of labor, and increase production efficiency, Industry 4.0 is pursued by companies.
The changes bring out by Industry 4.0 are far more than that. With unmanned services, such as unmanned supermarkets and unmanned driving, it has become a trend that the industry replaced most of its human resources with robots. The unmanned factory becomes one of the significant changes. Large scale factory can be monitored by just a few people but still conduct production every hour and moment, which reduce a huge amount of the cost of labor.
In this thesis, we aim to build a mobile robot system with machine vision and object detection based on ROS that includes indoor obstacle avoidance and path planning functionalities. With the three-dimension vision system, the mechanical arm process the object recognition and the grip function which made the mobile robot apply to the unmanned factory.
Using the three-dimensional point cloud as the input of machine vision has the advantage of performing a more accurate pose estimation than two-dimension vision. The result performs more accurately and rapidly with diverse 3D machine vision algorithms than a single one. Finally, the practicality of using algorithms and mechanical arms for pose estimation and pick-and-place classification of real objects will be verified.

第一章 緒論 1
1.1 前言 1
1.2 研究動機與目的 4
1.3 文獻回顧 7
1.4 論文架構 8
第二章 移動平台設計 9
2.1 前言 9
2.2 ROS概念與架構 9
2.3 運動控制 14
2.4 定位 18
2.5 光達建圖 19
2.6 導航 25
第三章 位姿估算演算法 31
3.1 前言 31
3.2 三維空間的剛體運動位姿表示式 32
3.3 點雲對特徵(PPF) 40
3.3.1 模型特徵描述 40
3.3.2 換算場景座標 42
3.3.3 投票方式 45
3.3.4 位姿分類 48
3.4 迭代最近點(ICP) 49
第四章 實驗設計與結果 63
4.1 前言 63
4.2 實驗環境 63
4.3 實驗流程 69
4.4 移動平台建圖 71
4.5 虛擬場景 73
4.5.1 PPF估算物體位姿 74
4.5.2 PPF搭配ICP估算物體位姿 89
4.6 真實場景 104
4.6.1 PPF和ICP估算物體位姿 105
4.6.2 PPF和ICP配合手臂夾取 113
第五章 結論與未來展望 119
5.1結論 119
5.2未來展望 121
參考文獻 123

[1] Ben Wu, Masa Chen,“工業4.0”,https://oosga.com/industry4-0/
[2] 陳良榕, “無人工廠人人搶”,天下雜誌551期, 2014,
[3] 黃靖萱“揭祕鴻海深圳關燈工廠，「這設備到月球都能生產」”,商業週刊, 2019,
[4] “什麼是AGV 無人搬運車？”,台灣塔奇恩科技有限公司, 2019,
[5] K. S. Arun, T. S. Huang, S. D. Blostein,“Least-Squares Fitting of Two 3-D Point Sets”,IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-9, pp. 698-700, 1987
[6] P.J. Besl, Neil D. McKay,“A method for registration of 3-D shapes”, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 14, pp. 239-256, 1992
[7] Y. Chen, G. Medioni,“Object modeling by registration of multiple range images” IEEE International Conference on Robotics and Automation, 1991
[8] Zhengyou Zhang,“Iterative point matching for registration of free-form curves and surfaces” International Journal of Computer Vision, vol. 13, pp.119-152, 1994
[9] Bertram Drost, Markus Ulrich, Nassir Navab, Slobodan Ilic“Model Globally, Match Locally: Efficient and Robust 3D Object Recognition” IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2010
[10] 林佳蓁“以二維影像特徵作為ICP初始估測之三維點雲對位” 國立雲林科技大學資訊工程系，學術論文，2015
[11] 林哲宇“使用影像特徵與點雲訊息建構物件模型”淡江大學機械與機電工程學系碩士班，學術論文，2019
[12] 郭曜嘉“點對特徵物體姿態估測結合多視角投票方案用於工業機器人隨機取物”國立台灣大學電機工程學研究所，學術論文，2019
[13] “About ROS”,ros.org
[14] “ROS/Concepts”,ros.org
[15] “differential_drive”,ros.org
[16] “tf2”, ros.org
[17] Wim Meeussen,“Coordinate Frames for Mobile Platforms”, 2010
[18] “Lidar”, Wikipedia
[19] 光速小子,“3D 感測技術：什麼是 飛時測距”,知識力, 2021,
[20] 陳舜鴻,“運用雷射雷達(LiDAR)感測功能提升智慧汽車效能”,車輛研究測試中心資訊108期, 2015
[21] kate cheng,“LiDAR(光達)介紹”, ROS新手挑戰AI&Data組 系列 第 7 篇, 2020
[22] 工研院資通所Orlando Habe, 吳俊賢, 陳泓翔, 吳韻宜, 高世偉,“無人載具室內建圖技術”, 2019
[23] 林妏霙, 黃克雄, 洪連輝“剛體 〈Rigid Body〉”, 2010
[24] ccjou,“四元數與三維空間旋轉”,線代啟示錄, 2014
[25] “齊次座標”,
https://openhome.cc/Gossip/ComputerGraphics/HomogeneousCoordinate.htm
[26] Zedník Pavel,“Detection and Localization of Texture-Less Objects in RGB-D Images” Diplomové práce - 13133, 2015
[27] “ICP（迭代最近點）演算法”,2019
[28] Tolga Birdal,“ICP Registration using Efficient Variants and Multi-Resolution Scheme”,MathWorks, 2015,
[29] “當3D視覺感知遇上微機電系統─精細之眼，讓AI看懂世界”,科學月刊461期, 2020