臺灣博碩士論文加值系統

運動科學的演進不斷推陳出新，科學訓練已成為國內外訓練的基石，透過科學
化的訓練能夠提升訓練效益並增進運動選手的表現，因此運動科學介入訓練已
為當今不可或缺的趨勢。在眾多競技項目中，舉重作為一項強度極高的競技運
動，其精準的技巧和姿勢控制對運動員的表現至關重要。然而，現有的檢測設
備可能存在著無法全面捕捉運動員動作細節的限制，但隨著深度學習在影像處
理領域的快速發展，我們有了更多應用於運動科學的可能性。因此，為了能更
好掌握訓練品質，本研究探討將深度學習與影像處理結合，以實現運動員姿態
分析。

本研究基於卷積神經網路的深度學習，提出的方法分為三個部分。首先，蒐集
各類場景和角度的試舉影像並進行標註，做為資料前處理。其次，以即時拍攝
方式或將錄製好的影像匯入檢測系統，並使用改良後的YOLOv8m-pose 模型作
為關鍵點檢測的主幹網路，用於舉重選手身體關鍵特徵點的提取。最後透過人
性化的人機介面(UI)利用人體各關鍵點，找出選手試舉時的關鍵角度、槓鈴軌
跡與動作位移速度。讓教練與選手更能掌握訓練計畫，這項研究為舉重項目提
供了一個有力的訓練輔助工具。

The evolution of sports science continues to bring forth innovations, and scientific
training has become the cornerstone of both domestic and international training regimens.
Through scientific training methods, the efficiency of training can be enhanced, thereby
improving the performance of athletes. Consequently, integrating sports science into
training has become an indispensable trend in today's athletic world. Among the various
competitive sports, weightlifting is a high-intensity sport where precise technique and
posture control are crucial to an athlete's performance. However, existing detection
equipment may have limitations in fully capturing the detailed movements of athletes.
With the rapid development of deep learning in image processing, we now have more
possibilities for applications in sports science. Therefore, to better monitor training quality,
this study explores the combination of deep learning and image processing to achieve
athlete posture analysis.

This study uses convolutional neural networks to present a deep learning method
divided into three parts. First, weightlifting videos from various scenarios and angles are
collected and annotated for data preprocessing. Second, using either real-time capture or
recorded videos, the improved YOLOv8m-pose model detects key body feature points of
weightlifters. Finally, a user-friendly human-computer interface (UI) uses these key
points to identify critical angles, barbell trajectories, and movement displacement speeds
during lifts. This helps coaches and athletes enhance training plans, offering a powerful
training aid for weightlifting.

目 錄
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研究背景與動機 1
1.2研究目的 1
1.3論文架構 2

第二章 文獻探討 3
2.1姿態檢測目的 3
2.2 近年運用於姿態檢測方式 4
2.2.1 槓鈴扣環感測器 4
2.2.2 Metric VBT 5
2.2.3 3D動作分析系統 6
2.2.4 基於深度學習於姿態檢測 7
2.3 現有姿態檢測網路架構 10
2.3.1 YOLOv8 10
2.3.2 OpenPose 12
2.3.3 RTMPose 13
2.3.4 比較現有姿態檢測網路架構 14

第三章 研究方法 15
3.1 研究方法概論 15
3.2 資料蒐集 17
3.3 資料前處理 19
3.3.1 資料集標註 19
3.3.2 資料集轉檔 21
3.4 姿態分析 22
3.4.1 關鍵點角度 22
3.4.2 槓鈴軌跡 24
3.4.3 動作位移速度 25
3.5 參數調整 25

第四章 研究結果與分析 26
4.1 實驗資料集 26
4.1.1 資料集增強 27
4.2 實驗環境 27
4.3 模型訓練 28
4.4 效能評估指標 28
4.5 不同檢測模型比較 29
4.6 人機介面設計 30

第五章 結論與未來展望 32
5.1結論 32
5.2 未來展望 32

參考文獻 33

[1] “Ultralytics YOLOv8 Docs Pose Estimation”. [Online]. Available:
https://docs.ultralytics.com/tasks/pose/.

[2] “Men's +109 kg Weightlifting | Tokyo Olympics Replays”. [Online]. Available:
https://www.youtube.com/watch?v=AQE7u2fYuxg&t=4145s.

[3] Higashi, S., et al., 2019, “Development of wearable EMG measurement system on
forearm for wrist gestures discrimination”, in 2019 IEEE 1st Global Conference on
Life Sciences and Technologies (LifeTech), pp. 250-251.

[4] 何維華, 林敬能, 蔚順華, 徐敬亭, 鍾寶弘, 2023, “精準運科通往奧運金牌之
路”, 人文與社會科學簡訊, 24卷2期.

[5] “Metric”. [Online]. Available: https://www.metric.coach/.

[6] 陳麒先, 陳韋翰, 葉明嘉, 劉強, 2021, “運動生物力學支援競技運動之做法”,
國民體育季刊, 208期.

[7] LeCun, Y., Bottou, L., Bengio, Y., Haffner, P., 1998, “Gradient-Based Learning
Applied to Document Recognition”, Proceedings of the lEEE, vol. 86, no. 11.

[8] “Ultralytics YOLOv8 Docs”. [Online]. Available: https://docs.ultralytics.com.

[9] Wei, S. E., Ramakrishna, V., Kanade, T., & Sheikh, Y. 2016. “Convolutional pose
machines”, In Proceedings of the IEEE conference on Computer Vision and Pattern
Recognition (pp. 4724-4732).

[10] Ke Li, Shijie Wang, Xiang Zhang, Yifan Xu, Weijian Xu, Zhuowen Tu.,2021,
“Pose Recognition with Cascade Transformers”, arXiv preprint arXiv: 2104.06976.

[11] “CMU-Perceptual-Computing-Lab/openpose”. [Online]. Available:
https://github.com/CMU-Perceptual-Computing-Lab/openpose.

[12] “RTMPose”. [Online]. Available: https://mmpose.readthedocs.io/en/latest/.

[13] “江大白| 深入淺出，YOLOv8演算法使用指南”. [Online]. Available:
https://blog.csdn.net/csdn_xmj/article/details/136313867.

[14] “open-mmlab/MMPose”. [Online]. Available:
https://github.com/open-mmlab/mmpose.

[15] “Sony α6600M”. [Online]. Available:
https://store.sony.com.tw/product/show/ff8080816cd70b2d016cff701988581e.

[16] “Logitech C310 HD”. [Online]. Available:
https://www.logitech.com/zh-tw/products/webcams/c310-hd-webcam.960
000631.html?sp=1&searchclick=logi.

[17] “Labelme”. [Online]. Available:
https://github.com/labelmeai/labelme.

[18] “OpenCV-Python 影像翻轉”. [Online]. Available:
https://blog.csdn.net/imxlw00/article/details/112787642.