臺灣博碩士論文加值系統

近年來由於疫情影響以及少子化愈加嚴重，機器取代人工處理勞動工作已經是時代的趨勢。現今大多數的自動化設備皆是單純載重使用，移動方式依靠軌道或是規劃道路線進行移動。為此本論文首先探討為自動化設備添加更多功能而非單純執行載重任務，如物件偵測等等。並探討對於物件偵測，如何根據企業需求使用客製化自定義小型資料集進行深度學習之物件偵測模型訓練。其次探討透過使用光學雷達之同時定位與地圖構建技術及Navigation2技術，在提升自動化設備移動路線彈性的同時具備控制及監控自動化設備的使用狀態。接著探討如何使用具有分散式通訊架構之第二代機器人操作系統建立車聯網系統，將自動化設備使用物件偵測時所需要運算的資料轉移至邊緣裝置。透過在邊緣裝置進行邊緣運算，以此降低自動化設備原先需要龐大運算資料的成本。最後探討使用上述技術建立之車聯網系統優缺點並針對自定義小型資料集之深度學習、邊緣運算等方面進行分析及提出改善方案。

In recent years, due to the impact of the pandemic and the increasing severity of declining birth rates, the trend of machines replacing manual labor has become prominent. Most of the current automation equipment is primarily used for simple load carrying tasks, and their movement relies on tracks or pre-planned routes. Therefore, this paper first explores adding more functionalities to automation equipment beyond simply performing load carrying tasks, such as object detection and more. It also discusses how to train object detection models using customized and small-scale datasets tailored to the specific needs of businesses through deep learning. Secondly, it explores the use of simultaneous localization and mapping techniques with optical radar, as well as Navigation2 technology, to enhance the flexibility of the movement routes of automated equipment while providing control and monitoring of their operational status. Next, it discusses how to establish a vehicle-to-everything system using a second-generation robot operating system with a distributed communication architecture. This system aims to transfer the computational data required for object detection during the operation of automated equipment to edge devices. By performing edge computing on edge devices, the aim is to reduce the cost of extensive computational data that was originally required by automated equipment. Finally, the advantages and disadvantages of the vehicle-to-everything system established using the above technologies are discussed, and the deep learning and edge computing of custom small data sets are analyzed and improved.

摘要…………………………i
Abstract…………………………ii
誌謝…………………………iii
目錄…………………………iv
表目錄…………………………vi
圖目錄…………………………vii
第一章 緒論…………………………1
1.1研究背景…………………………1
1.2研究動機與目的…………………………2
1.3論文架構…………………………3
第二章 文獻回顧…………………………4
2.1 Turtlebot…………………………4
2.1.1 Turtlebot3…………………………4
2.2 ROS…………………………5
2.2.1 ROS2…………………………6
2.2.2 DDS…………………………8
2.2.3發布及訂閱機制…………………………9
2.3 SLAM…………………………11
2.3.1 Cartographer…………………………11
2.3.2 Local 2D SLAM…………………………11
2.3.3 Closing Loops…………………………13
2.4 Navigation2…………………………15
2.5深度學習…………………………16
2.5.1卷積神經網路…………………………16
2.5.2遷移學習…………………………18
2.5.3 YOLO…………………………19
第三章 系統架構…………………………22
3.1實驗設備與環境…………………………22
3.2基於ROS2之車聯網系統…………………………25
3.2.1 車聯網通訊架構…………………………25
3.3基於YOLO模型之物件偵測方法…………………………26
3.4 RSU系統…………………………27
3.4.1物件偵測系統…………………………28
3.4.2自動導航系統…………………………28
3.5 OBU系統…………………………29
第四章 實驗與結果…………………………31
4.1相機標定…………………………31
4.2物件偵測模型分析…………………………33
4.2.1自定義資料集…………………………33
4.2.2圖像標籤…………………………34
4.2.3 YOLO物件偵測模型分析方法…………………………35
4.2.4 YOLO模型訓練結果之精確率分析…………………………36
4.2.5 OBU上YOLO模型辨識結果…………………………37
4.2.6影像主題壓縮比較…………………………43
4.3 SLAM地圖優化…………………………43
4.4 ROS2去中心化結構…………………………46
4.5物件偵測車聯網實踐…………………………50
4.5.1開啟OBU系統…………………………50
4.5.2開啟RSU系統…………………………51
4.5.3 A點至B點進行導航…………………………52
4.5.4 B點至A點進行導航…………………………53
4.5.5物件偵測結果…………………………55
第五章 結論與未來研究…………………………56
參考文獻…………………………57
Extended Abstract…………………………59

[1]Y. LeCun, Y. Bengio, and G. Hinton, "Deep learning," nature, vol. 521, no. 7553, pp. 436-444, 2015.
[2]M. R. Ghori, K. Z. Zamli, N. Quosthoni, M. Hisyam, and M. Montaser, "Vehicular ad-hoc network (VANET)," in 2018 IEEE international conference on innovative research and development (ICIRD), 2018: IEEE, pp. 1-6.
[3]F. Yang, S. Wang, J. Li, Z. Liu, and Q. Sun, "An overview of internet of vehicles," China communications, vol. 11, no. 10, pp. 1-15, 2014.
[4]H. Seo, K.-D. Lee, S. Yasukawa, Y. Peng, and P. Sartori, "LTE evolution for vehicle-to-everything services," IEEE communications magazine, vol. 54, no. 6, pp. 22-28, 2016.
[5]r. G. P. Project. "Study on LTE support for Vehicle-to-Everything (V2X) service." 3rd Generation Partnership Project. https://portal.3gpp.org/desktopmodules/Specifications/SpecificationDetails.aspx?specificationId=2898 (accessed May 3, 2023).
[6]ROBOTIS. "TurtleBot3." ROBOTIS. https://emanual.robotis.com/docs/en/platform/turtlebot3/features/#specifications (accessed February 26, 2023).
[7]O. Robotics. "ROS：Home." Open Robotics. https://www.ros.org/ (accessed February 26, 2023).
[8]O. Robotics. "Documentation - ROS Wiki." Open Robotics. http://wiki.ros.org/ (accessed February 26, 2023).
[9]Y. Maruyama, S. Kato, and T. Azumi, "Exploring the performance of ROS2," in 2016 International Conference on Embedded Software (EMSOFT), 2016: IEEE, pp. 1-10.
[10]O. Robotics. "Distributions — ROS 2 Documentation: Foxy documentation." Open Robotics. https://docs.ros.org/en/foxy/Releases.html (accessed February 26, 2023).
[11]G. Pardo-Castellote, "Omg data-distribution service: Architectural overview," in 23rd International Conference on Distributed Computing Systems Workshops, 2003. Proceedings., 2003: IEEE, pp. 200-206.
[12]C. Stachniss, J. J. Leonard, and S. Thrun, "Simultaneous localization and mapping," Springer Handbook of Robotics, pp. 1153-1176, 2016.
[13]W. Hess, D. Kohler, H. Rapp, and D. Andor, "Real-time loop closure in 2D LIDAR SLAM," in 2016 IEEE international conference on robotics and automation (ICRA), 2016: IEEE, pp. 1271-1278.
[14]S. Agarwal, Mierle, K., & The Ceres Solver Team. "Ceres Solver — A Large Scale Non-linear Optimization Library." Google Inc. http://ceres-solver.org/ (accessed April 10, 2023).
[15]K. Konolige, G. Grisetti, R. Kümmerle, W. Burgard, B. Limketkai, and R. Vincent, "Efficient sparse pose adjustment for 2D mapping," in 2010 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2010: IEEE, pp. 22-29.
[16]S. Macenski, F. Martín, R. White, and J. G. Clavero, "The marathon 2: A navigation system," in 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2020: IEEE, pp. 2718-2725.
[17]A. K. Jain, J. Mao, and K. M. Mohiuddin, "Artificial neural networks: A tutorial," Computer, vol. 29, no. 3, pp. 31-44, 1996.
[18]D. A. Forsyth et al., "Object recognition with gradient-based learning," Shape, contour and grouping in computer vision, pp. 319-345, 1999.
[19]S. Hochreiter and J. Schmidhuber, "Long short-term memory," Neural computation, vol. 9, no. 8, pp. 1735-1780, 1997.
[20]I. Goodfellow et al., "Generative adversarial networks," Communications of the ACM, vol. 63, no. 11, pp. 139-144, 2020.
[21]Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition," Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, 1998.
[22]S. J. Pan and Q. Yang, "A survey on transfer learning," IEEE Transactions on knowledge and data engineering, vol. 22, no. 10, pp. 1345-1359, 2010.
[23]S. Ruder, "An overview of multi-task learning in deep neural networks," arXiv preprint arXiv:1706.05098, 2017.
[24]J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You only look once: Unified, real-time object detection," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 779-788.
[25]C. Li et al., "YOLOv6: A single-stage object detection framework for industrial applications," arXiv preprint arXiv:2209.02976, 2022.
[26]C. Li et al., "YOLOv6 v3. 0: A Full-Scale Reloading," arXiv preprint arXiv:2301.05586, 2023.
[27]Ultralytics. "Ultralytics | Revolutionizing the World of Vision AI." Ultralytics. https://ultralytics.com/ (accessed May 5, 2023).
[28]C. Consortium. "COCO - Common Objects in Context." COCO Consortium. https://cocodataset.org/#home (accessed March 17, 2023).
[29]C.-Y. Wang, A. Bochkovskiy, and H.-Y. M. Liao, "YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors," in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 7464-7475.
[30]I. F. Akyildiz, X. Wang, and W. Wang, "Wireless mesh networks: a survey," Computer networks, vol. 47, no. 4, pp. 445-487, 2005.