臺灣博碩士論文加值系統

智慧製造的轉型過程中，取代基本的人員手動操作乃為優先探討的議題。使用直覺視覺判讀於重複取放工件是常見的人員手動操作，當轉換為自動操作時則須面對可能有不同款工件凌亂堆疊而產生之局部輪廓影像解讀問題，況且使用機器手臂之爪夾代勞時還有抓取方位上之各種限制，但這些現象都可藉由幾何演算求解。本研究乃使用深度相機攝得固定視角工作畫面之3D點雲為基礎，先將能夠被爪夾正確夾持之工件局部幾何定義為可夾持特徵，藉由攝得多視角可夾持特徵之點雲做深度學習之訓練。實作時比對攝得之工作區域點雲與各可夾持特徵。由於每個可夾持特徵亦同時定義了爪夾應到達點及近接向量，當由點雲比對識出可夾持特徵之3D方位後，再以距離與角度等約束條件篩選出機器手臂實際可達到之抓取特徵，即可提供機器手臂以逆向運動反推出關節動作來實作，完成取代人員抓取之操作。

During the transformation toward smart manufacturing, replacing basic manual operations with autonomous mechanism is of the top priority. A common task of pick-and-place with visual assists can be a challenge when converting into robotic operations, because it has to face the problem of workpart contour interpretation caused by unexpected stack pile. Moreover, restrictions may also come from grasp orientations and positions. Nevertheless, the geometric problem can be clearified by formulated calculations, as long as the feasible grasp fratures are identified. In this study, the 3D point cloud of the oriented part image is used as the training basis. The essential geometric portions of a workpiece that can be properly clamped by the gripper is defined as the grasp features. Then mutiple orientation point clouds of the grasp features are captured for deep learning training. In the implementation, the real time point cloud of the work area is iteratively compared with each grasp feature to identify grasp potential. Since every grasp feature will also has its approaching vector and grasp point associated, when a grasp feature is identified, the ideal vector and point are immediately acquired. The reachable grasp features are then recognized if the candidate features fit the physical constraints, and the joint movements can then be generated by inverse kinematic.

摘要...i
Abstract...ii
誌謝...iv
目錄...v
表目錄...vii
圖目錄...viii
第一章 緒論...1
1.1研究背景...1
1.2研究動機...2
1.3研究目的...2
1.4研究限制...3
1.5章節說明...3
第二章 文獻探討...4
2.1點雲特徵識別(Feature Recognition by Point Clouds)...4
2.1.特徵識別(Feature Recognition)...4
2.1.2 點雲應用(Point Cloud)...5
2.1.3點雲配準(Point Cloud registration)...5
2.2 機器手臂特徵抓取(Robot Grasp Feature)...6
2.3深度學習(Deep Learning)...7
2.4智慧製造(Smart Manufacturing)...9
2.5文獻總結...10
第三章 研究方法...12
3.1方法概述...12
3.2研究環境...15
3.3工件輪廓分析...17
3.3.1最適夾持點...18
3.3.2夾持特徵定義...18
3.4實驗點雲資料...24
3.4.1特徵區域點雲...25
3.4.2實驗之場景...26
3.5 深度學習...28
3.5.1 JSIS3D方法簡介...28
3.6迭代最近點(Iterative Close Point)...30
3.6.1 ICP目標函數...31
第四章 實作部分...32
4.1實作設備介紹...32
4.2實驗點雲資料獲取...33
4.2.1特徵區域點雲獲取...34
4.2.2 實驗場景點雲獲取...35
4.3 JSIS3D實作...37
4.4實作結果(對照圖片)...38
4.5結果分析...47
第五章 結論...48
5.1結論...48
5.2未來研究...48
參考文獻...49
Extended Abstract...52

Bauer, P., Heckler, L., Worack, M., Magaña, A., & Reinhart, G., 2021, Registration strategy of point clouds based on region-specific projections and virtual structures for robot-based inspection systems, Measurement, Vol. 185, pp. 109963.
Bu, L., Zhang, Y., Liu, H., Yuan, X., Guo, J., & Han, S., 2021, An IIoT-driven and AI-enabled framework for smart manufacturing system based on three-terminal collaborative platform, Advanced Engineering Informatics, Vol. 50, pp. 101370.
Chang, Q., & Xiong, Z., 2020, Vision-aware target recognition toward autonomous robot by Kinect sensors, Signal Processing: Image Communication, Vol. 84, pp. 115810.
Chen, L., Huang, P., & Meng, Z., 2019, Convolutional multi-grasp detection using grasp path for RGBD images, Robotics and Autonomous Systems, Vol. 113, pp. 94-103.
Han, X., Dong, Z., & Yang, B., 2021, A point-based deep learning network for semantic segmentation of MLS point clouds, ISPRS Journal of Photogrammetry and Remote Sensing, Vol. 175, pp.199-214.
Hietanen, A., Latokartano, J., Foi, A., Pieters, R., Kyrki, V., Lanz, M., & Kämäräinen, J. K., 2021, Benchmarking pose estimation for robot manipulation, Robotics and Autonomous Systems, Vol. 143, pp. 103810.
Le, T. T., Le, T. S., Chen, Y. R., Vidal, J., & Lin, C. Y., 2021, 6D pose estimation with combined deep learning and 3D vision techniques for a fast and accurate object grasping, Robotics and Autonomous Systems, Vol.
Lei, H., Jiang, G., & Quan, L., 2017, Fast descriptors and correspondence propagation for robust global point cloud registration, IEEE Transactions on Image Processing, Vol. 26, No.8, pp. 3614-3623.
141, pp. 103775.
Namjoshi, J., & Rawat, M., 2022, Role of smart manufacturing in industry 4.0, Materials Today: Proceedings.
Qi, S., Ning, X., Yang, G., Zhang, L., Long, P., Cai, W., & Li, W., 2021, Review of multi-view 3D object recognition methods based on deep learning, Displays, Vol. 69, pp. 102053.
Ramisa, A., Alenyà, G., Moreno-Noguer, F. & Torras, C, 2016, A 3D descriptor to detect task-oriented grasping points in clothing, Pattern Recognition, Vol. 60, pp. 936-948.
Rusu, R. B., Blodow, N., & Beetz, M., 2009, Fast point feature histograms (FPFH) for 3D registration, 2009 IEEE international conference on robotics and automation pp. 3212-3217.
Shi, P., Qi, Q., Qin, Y., Scott, P. J., & Jiang, X., 2022, Highly interacting machining feature recognition via small sample learning. Robotics and Computer-Integrated Manufacturing, Vol. 73, pp. 102260.
Song, Y., He, F., Duan, Y., Liang, Y., & Yan, X., 2022, A Kernel Correlation-Based Approach to Adaptively Acquire Local Features for Learning 3D Point Clouds, Computer-Aided Design, Vol. 146, pp. 103196.
Srivastava, S., & Lall, B., 2019, Deeppoint3d: Learning discriminative local descriptors using deep metric learning on 3d point clouds, Pattern Recognition Letters, Vol. 127, pp. 27-36.
Tao, Y., & Zhou, J., 2017, Automatic apple recognition based on the fusion of color and 3D feature for robotic fruit picking, Computers and Electronics in Agriculture, Vol. 142, pp. 388-396.
Wei, P., Yan, L., Xie, H., & Huang, M., 2022, Automatic coarse registration of point clouds using plane contour shape descriptor and topological graph voting, Automation in Construction, Vol. 134, pp. 104055.
Xin, M., Li, B., Wei, X., & Zhao, Z., 2021, Rapid registration method by using partial 3D point clouds, Optik, Vol. 246, pp. 167764.
Yang, L., Liu, Y., Peng, J., & Liang, Z., 2020, A novel system for off-line 3D seam extraction and path planning based on point cloud segmentation for arc welding robot, Robotics and Computer-Integrated Manufacturing, Vol.64, pp. 101929.
Yang, Y., Tang, R., Wang, J., & Xia, M., 2021, A hierarchical deep neural network with iterative features for semantic labeling of airborne LiDAR point clouds, Computers & Geosciences, Vol.157, pp. 104932.
Zhou, Z., Li, L., Fürsterling, A., Durocher, H. J., Mouridsen, J., & Zhang, X., 2022, Learning-based object detection and localization for a mobile robot manipulator in SME production, Robotics and Computer-Integrated Manufacturing, Vol. 73, pp. 102229.