臺灣博碩士論文加值系統

鏈球投擲在田徑比賽中是一項技術動作較為複雜的運動項目，對運動員的協調性及
力量的要求很高，運動員需利用重心移動來帶動鏈球的加速，在旋轉過程中如果身體沒
有足夠的力量去抵抗球的旋轉力，在投擲時身體會變得非常不穩定，進而影響選手投擲
的距離。
隨著科技的進步許多運動項目都導入運動科學，
但 導入運動科學往往需要花費非常
多費用 且量測 時也需要穿著整套的設備， 如果 在烈日下訓練可能會影響選手的狀態 ，所
以 希望建立一套較其他分析設備便宜且選手不用穿著任何設備的情況下去分析選手的
運動狀態 。
本研究與國立虎尾科技大學的田徑隊合作，透過空拍機與
Gopro拍攝選手的運動過
程。空拍的影像會透過 Yolov4演算法進行頭部以及鏈球的辨識，以觀察選手在運動過程
的穩定度及旋轉過程中鏈球的速度變化。 Gopro拍攝的影像 會透過 MoveNet分析選手身
上的關節點位，透過這些點位去計算選手在運動過程中的 關節 角度 及高度的 變化， 在後
再 利用 Excel的樞紐分析圖將資料呈現出來，讓教練馬上就能知道， 以協助教練的調整
選手的課表，以達到訓練最佳化的成果。

Hammer throwing is a sport with complex technical movements in track and field competition. High demands on athlete’s coordination and strength. The athlete needs to use the center of gravity to drive the acceleration of the hammer. If the body does not have enough strength to resist the rotational force of the ball during the rotation, the body will become very unstable when throwing. it affects the throwing distance of the athlete.
With the advancement of technology, more and more sports are imported sports science. However, the import of sports science often costs a lot of money and requires wearing a full set of equipment when measuring. If training in the hot sun may affect the state of the athletes. So I hope to build a set of analysis equipment that is cheaper than other analysis equipment and the player does not need to wear any equipment to analyze the player's sports state.
This research is in collaboration with the track and field team of National Formosa University. Shoot the athletes movement process through the aerial camera and Gopro. The aerial images will be identified by the Yolov4 algorithm for the head and hammer. In order to observe the stability of the athlete during the movement and the speed change of the hammer during the rotation process. The images captured by Gopro will analyze the joints of the athletes through MoveNet. These points are used to calculate the angle changes of the athletes during the exercise. It can assist the coach to adjust the athlete schedule to achieve the best training effect.

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......1
1.3 論文架構......1
第二章 文獻探討......3
2.1 鏈球運動......3
2.2 運動科學......3
2.3 人工智慧(Artificial Intelligence, AI)......3
2.3.1 機器學習(Machine Learning)......4
2.3.2 深度學習(Deep Learning)......4
2.3.3 卷積神經網路(CNN)......4
2.3.4 特徵金字塔(Feature Pyramid Networks, FPN)......6
2.4 骨幹分析 (Pose Estimation)......7
2.4.1 MoveNet......8
2.5 Yolo......9
第三章 研究方法與實作......12
3.1 研究架構......12
3.1.1 影像拍攝......13
3.1.2 Yolo模型訓練......13
3.1.3 影像分析及資料可視化......14
3.2 影像收集......15
3.3 影像分析......17
3.3.1 各關節角度與高度之計算......17
3.3.2 選手運動軌跡追蹤......19
第四章 系統設計與成果......20
4.1 研究設備......20
4.1.1 側拍相機Gopro Hero 5......20
4.1.2 空拍機DJI Mavic Air 2......20
4.1.3 Yolo模型訓練主機......20
4.1.4 研究設備架設......21
4.2 系統架構......21
4.3 系統實作......22
4.3.1 Yolo模型訓練......22
4.3.2 Yolo參數檔設定......23
4.3.3 空拍影片分析......23
4.3.4 側拍影片分析......24
4.3.5 資料表格式......25
4.3.6 資料視覺化......25
4.4 研究成果......26
4.4.1 手臂與垂直軸之間的角度曲線......26
4.4.2 肩膀與手腕的高度差曲線......26
4.4.3 左右腳膝蓋角度曲線......27
4.4.4 左右腳踝與髖關節高度差曲線......27
4.4.5 右手肘彎曲角度曲線......27
4.4.6 鏈球速度曲線......28
4.4.7 選手運動軌跡......28
第五章 結論與未來發展......29
5.1 結論......29
5.2 未來發展......29
參考文獻......30
Extended Abstract......32

[1] Judge, L. W., Hunter, I., & Gilreath, E. (2008). Using sport science to improve coaching: A case study of the American record holder in the women's hammer throw. International Journal of Sports Science & Coaching, 3(4), 477-488.
[2] 陳文良 與 林宗正 ，，( 鏈球投 擲動作之定性分析 中華體育季刊， ，10( 65-73。
[3] 吳淑真 ( 臺灣體育運動科學發展現況與未來趨勢 體育學報 49( 1-14。
[4] 李文姬 ，，( 木球之運動科學應用與技術之探討 大專體育 ，，( 22-28。
[5] 劉錦章、黃長福、侯金賢（ 2000）。鏈球投擲之運動學分析。大專體育 ，30 1-9。
[6] Bolotin, A. E., Bakayev, V. V., Orlova, N., & Kozulko, A. (2017). Peculiarities of time structure and of biomechanical organization of a construction of motor actions in the hammer throw. In 8th INTERNATIONAL SCIENTIFIC CONFERENCE ON KINESIOLOGY (pp. 137-141).
[7] Judge, L. W., Hunter, I., & Gilreath, E. (2008). Using sport science to improve coaching: A case study of the American record holder in the women's hammer throw. International Journal of Sports Science & Coaching, 3(4), 477-488.
[8] 周昱百 ，，( 人工智慧於視網膜疾病之應用 臨床醫學月刊 , 86(1), 407-409。
[9] 盧羿程 、 張瑞益 ，，( 應用機器學習於預測維護診斷之馬達故障頻譜研究 中
國造船暨輪機工程學刊 38(3& 157-163。
[10] Bengio, Y. (2009). Learning deep architectures for AI. Now Publishers Inc.
[11] Schmidhuber, J. (2015). Deep learning in neural networks: An overview. Neural networks, 61, 85-117.
[12] O'Shea, K., & Nash, R. (2015). An introduction to convolutional neural networks. arXiv preprint arXiv:1511.08458.
[13] 蔡俊明與林育正。 (2019)。使用深度學習方法作大量資料的商品辨識。 TANET2019 臺灣網際網路研討會， 193-198。
[14] Fukushima, K. (1980). A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biol. Cybern., 36, 193-202.
[15] Jin, S., Ma, X., Han, Z., Wu, Y., Yang, W., Liu, W., ... & Ouyang, W. (2017). Towards multi-person pose tracking: Bottom-up and top-down methods. In ICCV PoseTrack Workshop (Vol. 2, No. 3, p. 7).
[16] Lin, T. Y., Dollár, P., Girshick, R., He, K., Hariharan, B., & Belongie, S. (2017). Feature pyramid networks for object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 2117-2125).
[17] Bajpai, R., & Joshi, D. (2021). Movenet: A deep neural network for joint profile prediction across variable walking speeds and slopes. IEEE Transactions on Instrumentation and Measurement, 70, 1-11.
30
[18] Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016). You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 779-788).