簡易檢索 / 詳目顯示

回結果列表
研究生： 謝御琦
Hsieh, Yu-Chi
論文名稱： 三維棧板孔洞座標估計之辨識方法研究
A Study of Recognition Methods for 3D Pallet Hole Coordinate Estimation
指導教授： 侯廷偉
Hou, Ting-Wei
學位類別： 碩士
Master
系所名稱： 工學院 - 工程科學系
Department of Engineering Science
論文出版年： 2026
畢業學年度： 114
語文別： 中文
論文頁數： 78
中文關鍵詞： 棧板辨識 、姿態估計 、深度學習 、PnP姿態解算 、點雲幾何推理
外文關鍵詞： Pallet Recognition, Pose Estimation, Deep Learning, PnP Pose Solving, Point Cloud Geometric Reasoning
相關次數： 點閱：4  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 隨著倉儲物流自動化需求提升，自主移動機器人（AMR）於搬運任務中之棧板辨識與對接定位精準度，攸關作業成功率與安全性。為降低部署成本並提升規格通用性與環境抗擾性，本研究以ToF相機同步取得深度圖與RGB影像，提出三種棧板孔洞中心三維座標推估架構，統一輸出左右孔洞中心座標，並建立可重現之評估流程。方法一採用YOLOv8n邊界框偵測，結合深度資訊融合與平面擬合校準，以快速定位孔洞區域並回推三維座標；方法二為純幾何點雲先驗約束式推估，透過前處理、先驗約束式切割與模型資料庫匹配進行座標推算；方法三為YOLOv8n-Pose關鍵點估測結合 PnP 姿態解算與多規格先驗模型，以提升多規格擴充性。

    實驗涵蓋正常光線與角度偏移、背光深色（含黑色及彩色棧板）、複雜貨架背景及牙叉遮擋等情境，並以成功率、深度誤差、孔距誤差與推論時間量化比較。結果顯示三種方法之推估成功率皆為100 %；在深度(Z軸)精度方面，方法二MAE_Z為22.33 mm，優於方法一（64.00 mm）與方法三（61.00 mm）。在孔洞間距誤差方面，方法二與方法三分別為1.0 mm與2.0 mm，方法一為10.5 mm。推論時間方面，方法一、二、三分別為293 ms、327 ms與428 ms。綜合而言，三個研究方法皆具可用性，若以幾何一致性與深度估測準確性為優先，且可取得品質穩定之深度點雲，本研究提出之方法二為主要部署方案；方法一適合強調即時性與精度容忍度較高之情境；方法三則在低成本考量、多規格擴充與姿態資訊需求下具備應用彈性。

    Warehouse automation increasingly relies on precise pallet recognition and docking localization for Autonomous Mobile Robots (AMRs), where estimation errors can jeopardize operational safety and task success. To improve environmental robustness and specification versatility while keeping deployment cost low, this thesis utilizes a Time-of-Flight (ToF) camera to synchronously capture depth maps and RGB data, and proposes three pipelines to estimate the 3D coordinates of left and right pallet-hole centers via a unified output interface. Method 1 leverages YOLOv8n bounding-box detection integrated with RANSAC-based plane calibration for rapid 3D back-projection. Method 2 employs robust geometric reasoning with adaptive point-cloud segmentation and model-database matching. Method 3 fuses YOLOv8n-Pose keypoint estimation with Perspective-n-Point (PnP) pose solving and multi-specification priors to maximize extensibility. Experiments on a fixed test set across five challenging scenarios—including angular offsets, backlit dark/colored pallets, complex rack backgrounds, and fork occlusion—show a 100% success rate for all methods. Method 2 achieves superior depth accuracy (MAE_Z = 22.33 mm) and geometric consistency, whereas Method 1 offers the highest efficiency (293 ms). Overall, Method 2 is recommended for high-precision requirements, while Methods 1 and 3 are viable alternatives for latency-sensitive or pose-aware applications, respectively.

    摘要 I Extended Abstract II 致謝 V 目錄 VI 表目錄 IX 圖目錄 X 第1章 緒論 1 1.1 研究動機 1 1.2 研究目的 3 1.3 研究貢獻 3 1.4 章節概要 4 第2章 文獻探討 5 2.1 基於3D幾何特徵與傳統機器視覺之技術 5 2.2 深度學習與深度資訊融合之技術 6 2.3 關鍵點偵測與姿態估計之技術 7 2.4 深度感測器比較 8 2.4.1 飛行時間法（Time-of-Flight, ToF） 8 2.4.2 立體視覺技術（Stereo Vision） 9 2.5 針孔成像模型（Pinhole Camera Model） 10 2.5.1 針孔成像原理 10 2.5.2 四大座標系之轉換關係 11 2.5.3 透鏡畸變模型（Lens Distortion Model） 11 2.6 點雲處理與幾何分割理論 12 2.6.1 直方圖統計（Histogram Statistics） 12 2.6.2 RANSAC平面擬合 13 2.6.3 歐幾里得聚類（Euclidean Clustering） 13 2.6.4 Kd-Tree近鄰搜尋 15 2.7 YOLO系列即時目標偵測模型 16 2.7.1 Ultralytics YOLOv8n 16 2.7.2 Ultralytics YOLO11n 16 2.7.3 本研究採用 YOLOv8n與 YOLO11n之考量 17 2.8 YOLOv8n-Pose 17 2.9 Perspective-n-Point（PnP） 18 2.9.1 PnP問題之本質 18 2.9.2 常用求解演算法分析 18 第3章 實驗架構 19 3.1 硬體設備 19 3.1.1 ToF相機 19 3.1.2 工業電腦IPC 21 3.1.3 相機安裝位置 21 3.2 軟體環境與開發工具 22 3.3 方法一: YOLOv8n邊界框偵測與深度推估方法 23 3.3.1 技術路線概述 23 3.3.2 訓練資料集建置 23 3.3.3 YOLOv8n邊界框提取 25 3.3.4 座標映射與中心點提取 25 3.3.5 深度主層篩選 25 3.3.6 RANSAC平面擬合 26 3.3.7 3D座標計算與驗證 26 3.4 方法二:純幾何點雲先驗約束式棧板孔洞推估方法 27 3.4.1 技術路線概述 27 3.4.2 點雲預處理 28 3.4.3 先驗(priori)約束式棧板切割 29 3.4.4 棧板腿部聚類演算法 32 3.4.5 模型資料庫匹配技術 33 3.4.6 姿態估計與座標轉換 34 3.5 方法三: YOLOv8n-Pose關鍵點姿態估測與幾何約束 36 3.5.1 技術路線概述 36 3.5.2 訓練資料集建置 37 3.5.3 YOLOv8n關鍵點偵測 38 3.5.4 棧板先驗知識（3D棧板模型建立） 38 3.5.5 SolvePnP迭代求解 39 3.5.6 幾何一致性 40 3.5.7 孔洞3D座標計算 41 3.6 實驗設計與效能評估 42 3.6.1 實驗設計 42 3.6.2 建立評估指標 43 第4章 實驗結果與討論 46 4.1 三種方法之整體性能比較 46 4.1.1 深度學習模型訓練結果 46 4.1.2 孔洞間距誤差評估 48 4.1.3 深度估測誤差比較 52 4.1.4 算效率比較 53 4.1.5 整體成功率統計 54 4.2 各測試情境下的性能分析 57 4.2.1 正常光線場景 58 4.2.2 背光環境（含黑色及其他彩色棧板） 58 4.2.3 貨架場景（背景複雜） 59 4.2.4 模擬相機安裝於叉車上 59 4.3 總結 60 第5章 結論與未來研究方向 61 5.1 結論 61 5.2 未來研究方向 62 參考文獻 63

    [1] W. Chen, W. Chi, S. Ji, H. Ye, J. Liu, Y. Jia, J. Yu, and J. Cheng, “A Survey of Autonomous Robots and Multi-robot Navigation: Perception, Planning and Collaboration,” Biomimetic Intelligence and Robotics, Vol. 5, No. 2, Art. no. 100203, 2025, doi: 10.1016/j.birob.2024.100203.
    [2] J. Huemer, M. Murschitz, M. Schörghuber, L. Reisinger, T. Kadiofsky, C. Weidinger, M. Niedermeyer, B. Widy, M. Zeilinger, C. Beleznai, T. Glück, A. Kugi, and P. Zips, “ADAPT: An Autonomous Forklift for Construction Site Operation,” arXiv preprint arXiv:2503.14331, Mar. 2025, [Online]. Available: https://arxiv.org/abs/2503.14331, last accessed Jan. 21, 2026.
    [3] D. Caldana, et al., “Pallet and Pocket Detection Based on Deep Learning Techniques,” in Proc. IEEE International Conference on Robotics and Biomimetics （ROBIO）, 2024, doi: 10.1109/ROBOT61475.2024.10797410.
    [4] K. Kita and T. Kato, “Approach and Fork Insertion to Target Pallet Based on Image Measurement Method,” Sensors, Vol. 26, No. 1, Art. no. 154, 2026, doi: 10.3390/s26010154.
    [5] W. Wang, M. Shan, Y. Yue, and D. Wang, “Autonomous Target Docking of Nonholonomic Mobile Robots Using Relative Pose Measurements,” IEEE Transactions on Industrial Electronics, Vol. 68, No. 8, pp. 7233–7243, 2021, doi: 10.1109/TIE.2020.3001805.
    [6] E. Tsiogas, I. Kleitsiotis, I. Kostavelis, et al., “Pallet Detection and Docking Strategy for Autonomous Pallet Truck AGV Operation,” in Proc. IEEE/RSJ International Conference on Intelligent Robots and Systems （IROS）, 2021, pp. 3444–3451, doi: 10.1109/IROS51168.2021.9636270.
    [7] J. Nygårds, T. Hogström, and A. Wernersson, “Docking to Pallets with Feedback from a Sheet-of-light Range Camera,” Proceedings of the 2000 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2000), vol. 3, pp. 1853–1859, 2000, doi:10.1109/IROS.2000.895241.
    [8] B. Molter and J. Fottner, “Real-Time Pallet Localization with 3D Camera Technology for Forklifts in Logistic Environments,” in Proc. IEEE International Conference on Service Operations and Logistics, and Informatics （SOLI）, 2018, pp. 297–302, doi: 10.1109/SOLI.2018.8476740.
    [9] F. Weichert, S. Skibinski, J. Stenzel, et al., “Automated Detection of Euro Pallet Loads by Interpreting PMD Camera Depth Images,” Logistics Research, Vol. 6, No. 2, pp. 99–118, 2013, doi: 10.1007/s12159-012-0095-8.
    [10] A. Hu, C. Xiang, and C. Kong, “Study on 3D Pallet Recognition and Localization Based on a Depth Camera,” Journal of Mechanical Science and Technology, Vol. 39, No. 12, pp. 7753–7762, 2025, doi: 10.1007/s12206-025-1135-z.
    [11] J. Gu, H. Song, C. Li, and S. Zhu, “Research on the Method of Universal Forklift Pallet Detection and Pose Estimation Based on Visual and 3D Point Cloud Fusion,” in Proc. 5th International Conference on Mechatronics Technology and Intelligent Manufacturing （ICMTIM）, 2024, pp. 388–394, doi: 10.1109/ICMTIM62047.2024.10629322.
    [12] Z. Xiao, Z. Wang, X. Zhang, and D. Qian, “Research on Visual Localization of the Pallet Relative to the AGV Based on Image Matching,” in Proc. 6th International Conference on Robotics, Intelligent Control and Artificial Intelligence （RICAI）, 2024, pp. 1097–1102, doi: 10.1109/RICAI64321.2024.10911877.
    [13] T. Zhao, J. Liu, and J. Zhou, “High Precision Pallet Detection and Location Algorithm for AGV,” in Proc. IEEE International Conference on Cybernetics and Intelligent Systems （CIS） and IEEE Conference on Robotics, Automation and Mechatronics （RAM）, 2023, pp. 170–173, doi: 10.1109/CIS-RAM55796.2023.10370758.
    [14] H. Wang, B. Cui, X. Wen, Y. Jiang, C. Gao, and Y. Tian, “Pallet Detection and Estimation with RGB-D Salient Feature Learning,” in Proc. China Automation Congress （CAC）, 2023, pp. 8914–8919, doi: 10.1109/CAC59555.2023.10451178.
    [15] Y.-Y. Li, X.-H. Chen, G.-Y. Ding, et al., “Pallet Detection and Localization with RGB Image and Depth Data Using Deep Learning Techniques,” in Proc. International Conference on Automation, Control and Robotics Engineering （CACRE）, 2021, pp. 306–310, doi: 10.1109/CACRE52464.2021.9501390.
    [16] M. Zaccaria, R. Monica, and J. Aleotti, “A Comparison of Deep Learning Models for Pallet Detection in Industrial Warehouses,” in Proc. IEEE International Conference on Intelligent Computer Communication and Processing （ICCP）, 2020, pp. 417–422, doi: 10.1109/ICCP51029.2020.9266168.
    [17] Y. Shao, Z. Fan, B. Zhu, J. Lu, and Y. Lang, “A Point Cloud Data-Driven Pallet Pose Estimation Method Using an Active Binocular Vision Sensor,” Sensors, Vol. 23, No. 3, Art. no. 1217, 2023, doi: 10.3390/s23031217.
    [18] A. Keser and P. Oğuz Ekim, “Otonom Forklift Ile Palet Tespiti ve Hassas Yanaşma / Pallet Detection and Docking with Autonomous Forklift,” in Proc. Innovations in Intelligent Systems and Applications Conference （ASYU）, 2024, pp. 1–5, doi: 10.1109/ASYU62119.2024.10756973.
    [19] Y. Hamad, M. A. Khan, and A. Mohamed, “Object Depth and Size Estimation Using Stereo-Vision and Integration with SLAM,” arXiv preprint arXiv:2409.07623, Sep. 2024, doi: 10.48550/arXiv.2409.07623.
    [20] T.-T. Han, M. T. Duc, and H. D. Tan, “A Lightweight Distance Estimation Method Using Pinhole Camera Geometry Model,” Measurement Science and Technology, Vol. 36, No. 4, Art. no. 045406, 2025, doi: 10.1088/1361-6501/adbf3b.
    [21] Y. Du, Z. Jiang, J. Tian, and X. Guan, “Modeling, Analysis, and Optimization of Random Error in Indirect Time-of-Flight Camera,” Optics Express, Vol. 33, No. 2, pp. 1983–1994, 2025, doi: 10.1364/OE.547731.
    [22] S. Kumar, S. Dhatrak, A. Shah, M. Gupta, and M. Imran, “A Simple, Efficient and Complete Software Pipeline for Constructing an RGB-D Camera Using RGB and ToF Depth Cameras,” in Proc. International Conference on Mechatronics and Robotics Engineering （ICMRE）, 2025, pp. 47–54, doi: 10.1109/ICMRE64970.2025.10976309.
    [23] B. Langmann, K. Hartmann, and O. Loffeld, “Depth Camera Technology Comparison and Performance Evaluation,” in Proc. 1st International Conference on Pattern Recognition Applications and Methods （ICPRAM）, 2012, pp. 438–444, doi: 10.5220/0003778304380444.
    [24] S. Hussmann, T. Ringbeck, and B. Hagebeuker, “A Performance Review of 3D TOF Vision Systems in Comparison to Stereo Vision Systems,” in A. Bhatti （ed.）, Stereo Vision, InTech, 2008, pp. 103–120, doi: 10.5772/5898.
    [25] M. Dhanda, B. A. Rogers, S. Hall, E. Dekoninck, and V. Dhokia, “Reviewing Human-Robot Collaboration in Manufacturing: Opportunities and Challenges in the Context of Industry 5.0,” Robotics and Computer-Integrated Manufacturing, Vol. 93, Art. no. 102937, 2025, doi: 10.1016/j.rcim.2024.102937.
    [26] A. Efendi, Y.-H. Shao, and C.-Y. Huang, “Technological Development and Optimization of Pushing and Grasping Functions in Robot Arms: A Review,” Measurement, Vol. 242, Art. no. 115729, 2025, doi: 10.1016/j.measurement.2024.115729.
    [27] R.-J. Hong, Y.-R. Li, M.-H. Hung, J.-W. Chang, and J. C. Hung, “Integrating Object Detection and Semantic Segmentation into Automated Pallet Forking and Picking System in AGV,” in J. C. Hung, N. Y. Yen, and J.-W. Chang （eds.）, Frontier Computing, Springer Nature, 2022, doi: 10.1007/978-981-16-8052-6_13.
    [28] R. Khanam and M. Hussain, “YOLOv11: An Overview of the Key Architectural Enhancements,” arXiv preprint arXiv:2410.17725, Oct. 2024, doi: 10.48550/arXiv.2410.17725.
    [29] Ultralytics, “Explore Ultralytics YOLOv8,” [Online]. Available: https://docs.ultralytics.com/models/yolov8/, last accessed Dec. 22, 2025.
    [30] Ultralytics, “Model YAML Configuration Guide,” [Online]. Available: https://docs.ultralytics.com/guides/model-yaml-config/, last accessed Dec. 22, 2025.
    [31] Ultralytics, “Ultralytics YOLO11,” [Online]. Available: https://docs.ultralytics.com/models/yolo11/, last accessed Dec. 25, 2025.
    [32] 徐家良,可適性結構演算法應用於輔助堆高機棧板檢測,碩士論文，淡江大學電機工程學系碩士班，2024
    [33] 黃子祐, 叉式移動機器人之自動棧板對接系統的設計與實現, 碩士論文，淡江大學電機工程學系碩士班，2024.
    [34] C. Xu, T. Guo, Y. Huang, and L. Cheng, “A Hough Voting-Based 2-Point RANSAC Solution to the Perspective-n-Point Problem,” IEEE Transactions on Image Processing, vol. 34, pp. 6838–6851, 2025,doi:10.1109/TIP.2025.3622336.
    [35] H. Yao, P. Han, J. Chen, et al., “MonOri: Orientation-Guided PnP for Monocular 3-D Object Detection,” IEEE Transactions on Neural Networks and Learning Systems, Vol. 36, No. 10, pp. 19068–19080, 2025, doi: 10.1109/TNNLS.2025.3577618.
    [36] C. Mok, I. Baek, Y. S. Cho, Y. Kim, and S. B. Kim, “Pallet Recognition with Multi-Task Learning for Automated Guided Vehicles,” Applied Sciences, vol. 11, no. 24, art. no. 11808, 2021, doi:10.3390/app112411808.
    [37] N. Kai, H. Yoshida, and T. Shibata, “Pallet Pose Estimation Based on Front Face Shot,” IEEE Access, vol. 13, pp. 37624–37631, 2025, doi:10.1109/ACCESS.2025.3538045.
    [38] Orbbec (奥比中光科技集團股份有限公司), “座標系统 (Femto Bolt 文檔),” [Online]. Available: https://www.orbbec.com.cn/index/Sdkdoc/info.html?cate=117&id=37, last accessed Dec. 4, 2025.

    QR CODE