混線生產合併類似或相關的產品生產線，不同原物料與在製品隨機出現，生產系統須能即時、自動改變製程設定，完成對應的製造工作。本研究針對混線生產之鞋底貼合工作，發展智慧型多軸噴膠運動規劃技術，在無夾治具使用的情況下，針對工作平台上隨機擺置之不同鞋底，自動產生對應的噴膠路徑。首先評估不同深度攝影機在應用環境中的效能表現，選擇適合的裝置與成像條件；接著透過相機校正程序，整合點膠機、深度相機與真實世界座標系。主要創新內容在於使用深度相機，快速進行鞋底的三維辨識與定位，對應的計算方法分為兩個階段，首先根據Linemod方法，由真實場景影像中決定鞋底尺寸與款式，並估計該鞋底在空間中的粗略姿態；接下來以此結果做為初始解，透過修剪式疊代最近點演算法，計算出精確的鞋底姿態。根據鞋底在工作平台上的擺置位置與方位資訊，即可將預先建立好的噴膠路徑，進行對應的空間座標轉換，驅動五軸運動機構，產生高效率、無碰撞的噴膠路徑，自動完成上膠的作業，最後透過實際噴膠測試，驗證膠運動規劃技術的可行性。

Mixed line production is a critical element to realize the idea of mass product customization. Raw materials of product variants randomly appearing in the production require quick and automatic adjustment of the production system to complete the manufacturing task. This research develops an intelligent motion planning technology based on depth sensing for multi-axis glue dispensing in the mixed production of shoe bottoms. The goal is to automatically generate the motion trajectory of the dispensing valve for a shoe bottom erratically placed on a work table without use of fixturing. First, we evaluate the performance of three different depth cameras in the real manufacturing environment from multiple aspects. The evaluation result helps determine the best equipment and the proper lighting condition. Next, calibration procedures are conducted to establish the transformation matrices among the coordinate systems involved in the motion planning, including the dispensing machine, the depth camera, and the real world. The essential function of the motion planning technology is to recognize the shoe bottom existing in the real scene and to estimate its pose in the space. A two-stage algorithm is proposed to implement this function. The Linemod method is first applied to determine the rough pose from the training data corresponding to a set of templates. A trimmed ICP (Interactive Closest Point) method subsequently finds the precise pose using the rough one as the initial solution. Pre-determined dispensing paths can be transformed into the right position around the work part with the estimated pose. The corresponding motion commands then drive the dispenser controller to finish the gluing task. Test results of real shoe bottoms validate the effectiveness of the developed motion planning technology.

摘要 I
Abstract II
目錄 III
圖目錄 V
表目錄 VII
第一章、緒論 1
1.1、研究背景 1
1.2、文獻探討 2
1.3、研究目的 4
第二章、系統簡述與架構 5
2.1、系統環境與描述 5
2.2、系統架構 7
2.3、資通訊工具 8
2.4、硬體工具 10
第三章、方法論述 13
3.1、深度相機校正與評估 13
3.1.1、確立機器位置 14
3.1.2、相機校正 14
3.1.3、點膠機座標轉換 18
3.1.4、測試數據比較 19
3.2、結構化點雲 20
3.3、Linemod疊代模型規劃 21
3.3.1、Linemod方法簡述 22
3.3.2、疊代最近點法 24
3.4、路徑生成 26
第四章、系統實作 28
4.1、實作成果 29
4.1.1、整合Linemod與ICP 29
4.1.2、錯誤累積補償 31
4.1.3、噴膠後運動處理 33
4.1.4、實際噴膠 35
4.2、誤差驗證 45
4.2.1、方法誤差 46
4.2.2、總誤差（補償後） 48
第五章、結論與未來展望 50
5.1、研究結論 50
5.2、未來研究方向 51
參考文獻 53

[1] 黎百加，擴增實境中基於機器編程示範之運動規劃，國立清華大學，碩士論文，(2017)。
[2] 劉宇望，擴增實境中示範學習之運動規劃人機界面，國立清華大學，碩士論文，(2018)。
[3] Canny, J. A computational approach to edge detection. In Readings in computer vision, (1987) 184-203.
[4] Perona, P., & Malik, J. Scale-space and edge detection using anisotropic diffusion. In Transactions on pattern analysis and machine intelligence, IEEE, 12(7), (1990) 629-639.
[5] Arkin, E. M., Chew, L. P., Huttenlocher, D. P., Kedem, K., & Mitchell, J. S. An efficiently computable metric for comparing polygonal shapes. In Transactions on pattern analysis and machine intelligence, IEEE,13(3), (1991) 210-216.
[6] Belongie, S., Malik, J., & Puzicha, J. Shape matching and object recognition using shape contexts. IEEE Transactions on Pattern Analysis & Machine Intelligence, 4, (2002) 509-522.
[7] Brogefors, G. Hierarchical chamfer matching: A parametric edge matching algorithm. IEEE Transactions on Pattern Analysis & Machine Intelligence, 6, (1988) 849-865.
[8] Lowe, D. G. Object recognition from local scale-invariant features. In ICCV, 99(2), (1999) 1150-1157.
[9] Bay, H., Tuytelaars, T., & Van Gool, L. Surf: Speeded up robust features. In European conference on computer vision, (2006) 404-417.
[10] Wohlhart, P., & Lepetit, V. Learning descriptors for object recognition and 3d pose estimation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, (2015) 3109-3118.
[11] Chetverikov, D., Svirko, D., Stepanov, D., & Krsek, P. The trimmed iterative closest point algorithm. In Object recognition supported by user interaction for service robots, IEEE, 3, (2002) 545-548.
[12] 钟允晖，面向制鞋喷胶的机器人运动控制研究及其系统开发，浙江工业大学，碩士論文，(2012)。
[13] 武传宇、贺磊盈、李秦川、胡旭东，基于CAD模型的鞋底喷胶轨迹生成方法，《计算机辅助设计与图形学学报》，(2008)。
[14] 贾明峰、胡国清、吕成志，基于图像处理的自动喷胶系统的研究，《制造业自动化》，(2017)。
[15] 贾明峰，基于机器视觉的机器人自动喷胶系统的研制与应用，华南理工大学，碩士論文，(2018)。
[16] Khatib, O. Real-time obstacle avoidance for manipulators and mobile robots. In Autonomous robot vehicles, (1986) 396-404.
[17] Fischler, M. A., & Bolles, R. C. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography. In Communications of the ACM, 24(6), (1981) 381-395.
[18] Wiley, W. C., & McLaren, I. H. Time‐of‐flight mass spectrometer with improved resolution. In Review of scientific instruments, 26(12), (1955) 1150-1157.
[19] Hinterstoisser, S., Holzer, S., Cagniart, C., Ilic, S., Konolige, K., Navab, N., & Lepetit, V. Multimodal templates for real-time detection of texture-less objects in heavily cluttered scenes. In 2011 international conference on computer vision, IEEE, (2011) 858-865.
[20] OpenCV https://opencv.org/
[21] PCL http://pointclouds.org/