臺灣博碩士論文加值系統

本研究致力於開發一個綜合了物體偵測、對象排序和運動控制的系統，旨在實現在自主移動車輛上的人體跟蹤和跟隨功能。我們首先通過自行收集和標註大量圖片，並對You Only Look Once深度學習模型進行改進，調整超參數以提高其對人體的識別準確性和穩健性。經過訓練後的模型能夠準確識別和定位攝像頭視野中出現的人體。
其次，我們著重於改進Simple Online and Realtime Tracking算法的參數設定，以優化對檢測到的人體的排序和跟蹤效果。我們提出了一系列改進方案，包括對運動模型和關聯分數的調整，以提高對多個人體的同時跟蹤的準確性和穩定性。這些改進使得系統能夠更好地應對人體之間的相互遮擋和運動複雜性，從而實現更精確的跟蹤效果。
接下來，我們將處理後的資訊通過機器人作業系統Robot Operating System傳送至自主移動車輛端。並根據從ROS接收到的訊息，對車輛進行適當的運動控制，以實現對檢測到的人體的跟隨。
總而言之，本研究通過對YOLO模型和SORT算法的改進，實現了對於人體的即時跟隨，並且提出的改進策略使系統具有更高的準確性和穩定性。這些技術和方法具有廣泛的應用價值，尤其在自主移動車輛和人機交互領域有著潛在的重要意義。

This study aims to develop a system that integrates object detection, object sorting, and motion control to enable human tracking and following functionality in autonomous vehicles. First, we collected and annotated a large number of images and improved the You Only Look Once deep learning model by adjusting its hyperparameters to enhance its accuracy and robustness in human recognition. The trained model can accurately identify and locate humans within the camera's field of view.
Next, we focused on optimizing the parameter settings of the Simple Online and Realtime Tracking algorithm to improve the sorting and tracking of detected humans. We proposed a series of improvements, including adjustments to the motion model and association scores, to enhance the accuracy and stability of tracking multiple humans simultaneously. These enhancements enable the system to better handle occlusions and the complexity of human movement, resulting in more precise tracking.
Subsequently, we transmitted the processed information to the autonomous vehicle via the Robot Operating System. Based on the messages received from ROS, the vehicle performs appropriate motion control to follow the detected humans.
In summary, this study achieved real-time human following through improvements to the YOLO model and the SORT algorithm. The proposed enhancement strategies provide higher accuracy and stability to the system. These techniques and methods have broad application value, particularly in the fields of autonomous vehicles and human-computer interaction.

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研究動機及目的 1
1.2文獻探討 2
1.3 論文章節架構 4
第二章 相關理論與先前技術 5
2.1 YOLO(You Only Look Once) 5
2.2機器人作業系統 16
2.2.1分布式計算 17
2.2.2 封裝和重用 17
2.2.3通信機制 18
2.3 SORT算法 30
2.3.1 預測步驟 30
2.3.2關聯步驟 30
第三章 硬體架構介紹 32
3.1硬體基本架構 32
3.2 電控系統 38
3.2.1控制器 38
3.2.2馬達與驅動器 42
3.2.3 RS485通訊 48
第四章 實驗設計及結果 51
4.1 YOLO模型訓練 52
4.2使用之程式細節及參數調教 63
4.3實體車建構發想 68
4.4整合展示以及實驗結果 68
4.4.1整合展示 68
4.4.2實驗結果 72
第五章 總結與未來展望 74
5.1總結 74
5.2未來展望 75
參考文獻 76

[1]Farej, Z. K., & Ismail, A. N, “On the Location Accuracy of the Multi-Antenna GPS,” IEEE International Conference on Engineering, Science and Advanced Technology, pp. 29-33, June 2023.
[2]Li, C. H., Lao, K. W., & Tam, K. W, ”A flooding warning system based on RFID tag array for energy facility,”IEEE International Conference on RFID Technology & Application, pp. 1-4, September 2018.
[3]Anish, A., Sharan, R., Malini, A. H., & Archana, T, “ Enhancing Surveillance Systems with YOLO Algorithm for Real-Time Object Detection and Tracking,”IEEE International Conference on Automation, Computing and Renewable Systems, pp. 1254-1257, December 2023.
[4]Liu, K. C., Shen, Y. T., & Chen, L. G., “Simple online and realtime tracking with spherical panoramic camera,” IEEE International Conference on Consumer Electronics, pp. 1-6, January 2018.
[5]Plastiras, G., Kyrkou, C., & Theocharides, T., “You only look once: Unified, real-time object detection,” ACM Int. Conf. Proceeding Ser., 2018.
[6]Bochkovskiy, A., Wang, C. Y., & Liao, H. Y. M., “ Yolov4: Optimal speed and accuracy of object detection,”arXiv preprint arXiv:2004.10934., 2020.
[7]Alahi, A., Goel, K., Ramanathan, V., Robicquet, A., Fei-Fei, L., & Savarese, S., “Social lstm: Human trajectory prediction in crowded spaces,”IEEE conference on computer vision and pattern recognition, pp. 961-971, 2016.
[8]Sadat, A., Baumann, S., & Stiller, C., “ Predictive driving policy for socially-aware robot navigation among pedestrians,”IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 7072-7079, 2018.

[9]Ultralytics. https://docs.ultralytics.com/yolov5/tutorials/train_custom_data/
[10]Kaggle.
https://www.kaggle.com/datasets/meeratif/yolo-format-data
[11]Roboflow.
https://public.roboflow.com/
[12]Comet and YOLOv5.Ipynb.
https://colab.research.google.com/drive/1RG0WOQyxlDlo5Km8GogJpIEJlg_5lyYO?usp=sharing
[13]Ultralytics.
https://docs.ultralytics.com/yolov5/tutorials/model_export/
[14]ROS入門教學. <分布式通信.>
https://book.guyuehome.com/ROS2/2.%E6%A0%B8%E5%BF%83%E6%A6%82%E5%BF%B5/2.9_%E5%88%86%E5%B8%83%E5%BC%8F%E9%80%9A%E4%BF%A1/
[15]CSDN. <深入理解YOLOv5目标检测模型.> https://blog.csdn.net/qq_48322366/article/details/124904055#/ (2022, May 9).
[16]IThome. <深入淺出ROS 2 (2).>
https://ithelp.ithome.com.tw/articles/10216423 (2021, September 30).
[17]ROS Wiki. https://wiki.ros.org/ROS/Tutorials/UnderstandingTopics#/
[18]ROS Wiki.
https://wiki.ros.org/rviz
[19]Gazebo.
https://classic.gazebosim.org/tutorials
[20]ROS Wiki.
https://wiki.ros.org/rosbag
[21]陳宣珽，「ROS機器人教材之開發」，碩士論文，國立雲林科技大學電機工程系 , July 2019.
[22]Intel RealSense. https://www.intelrealsense.com/depth-camera-d435i#/
[23]Advantech. https://www.mouser.tw/ProductDetail/Advantech/MIO-2263JH-U0A1E?qs=CrD54wQY5XBvFchWqjH1pA%3D%3D
[24]工控之家. <產品詳细信息. 工控之家. > https://www.gkzhan.com/st158500/product_8569488.html#/
[25]PyTorch.
https://pytorch.org/get-started/locally/
[26]Git.
https://git-scm.com/download/win/
[27]HumanSignal.
https://github.com/HumanSignal/labelImg
[28]Haroon Shakeel. https://github.com/haroonshakeel/yolov7-object-tracking
[29]WongKinYiu.
https://github.com/WongKinYiu/yolov7
[30]ROS. https://docs.ros.org/en/melodic/api/std_msgs/html/index-msg.html
[31]CSDN. https://blog.csdn.net/weixin_43694096/article/details/124378167 (2022, April 25)
[32]ROS. https://docs.ros.org/en/noetic/api/geometry_msgs/html/msg/Twist.html