現今科學的進步推動了運動產業的快速發展，各項體育賽事的成績有著明顯的進步，相對來說就是加強各選手之間強烈的競爭意識，想要在各大賽事中嶄露頭角就必須有更效率與更專業的訓練方法，而運動科學就在其中扮演了非常重要的角色，但一套專業的運運動科學系統需要的費用是非常龐大的，普通的民眾甚至運動員都是無法承受的。因此，本研究透過影像辨識與大數據分析技術來發展一套運動科學分析系統。
以往運動選手在訓練時只能以教練口頭說明來進行調整，若將運動科學分析系統用於實際的訓練環境中，以視覺化及科學的方法幫助田徑隊在練習的過程中進行改善，讓選手清楚的了解自己的實際訓練狀況，降低因只能以想像訓練甚至溝通不良而產生的訓練效率下降，進而提升運動表現與競賽成績。
本研究使用Yolo影像辨識演算法與OpenCV影像處理技術，並與虎尾科技大學田徑隊進行合作，利用無人空拍機與Gopro運動相機拍攝選手奔跑影像，再透過影像的辨識與分析，計算出選手於奔跑時的頭部軌跡與速度變化，以觀察頭部軌跡與速度對訓練效果之影響；本研究亦使用MoveNet人體姿態演算法，透過辨識人體各關節點位，並計算各部位角度變化，同樣來判斷各部位角度變化對選手訓練效果之間的影響。
本研究結合大數據技術，將分析後的所有數據儲存下來，透過長時間的紀錄來進行訓練追蹤，且可依照影像、時間、選手來分類並選擇欲觀看之訓練結果。長時間的訓練紀錄可讓選手與教練清楚的知道各個階段的訓練品質與成效，數據化的呈現能讓訓練時的準確性有更好提升更大大的提升訓練效率。

Nowadays, the progress of science has promoted the development of the sports industry, various sports events’s results have significant progress. Relatively speaking, it is to increase the strong sense of competition between the players, if you want to emerge in the major events, you must have more efficient and professional training methods, and sports science plays a very important role in it, but a set of professional sports science system are costly, normal people and athletes can’t afford. Therefore, this research develops a motion science analysis system by image recognition and big data analysis technology.
In the past, athletes could use the coach’s verbal instructions to make adjustments during training. If the sports science analysis system is used in the actual training environment, it will help the track and field team to improve the skill with visual and scientific methods. They can better understand their actual training situation, prevent the reduced training efficiency caused by imaginary training or poor communication.
This research use Yolo algorithm and OpenCV, cooperates with the track and field team of National Formosa University. Using aerial camera and Gopro camera to record the running video of the athletes, then calculate the head trajectory and speed changes of the athletes while running by identification and analysis of images; This research also use the MoveNet human posture algorithm to identify the joint points of the human body and calculate the angle change. Determine the influence of the angle change of each part on the training effect.
This research combines big data technology to store all data after analysis, and track training through long-term records, it can classify and select training results to be viewed according to images, time, and players. Long-term training records allow athletes and coach to clearly understand the quality and effectiveness of training. Data presentation can improve the accuracy of training and greatly improve training efficiency.

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 研究流程.........................1
第二章 文獻探討.........................3
2.1 體育運動科學.....................3
2.2 影像與資料處理...................3
2.3 大數據技術.......................3
2.4 YOLO視覺辨識演算法...............4
2.5 人體姿態分析演算法(MoveNet).......7
第三章 研究方法與實作..................10
3.1 研究架構........................10
3.2 資料蒐集與解析...................13
3.3 MoveNet演算法...................14
3.4 Yolo演算法......................15
3.4.1 模型訓練........................15
第四章 系統設計與實作...................17
4.1 研究設備........................17
4.1.1 側拍相機Gopro Hero 5............17
4.1.2 空拍相機DJI Mavic Air 2.........17
4.1.3 訓練主機........................17
4.2 系統實作........................18
4.2.1 資料庫..........................18
4.2.2 資料蒐集........................19
4.2.3 分析流程.........................21
4.3 Yolo演算法模型...................26
4.3.1 模型訓練.........................26
4.3.2 參數設定........................26
4.4 視覺化呈現.......................27
4.4.1 側拍畫面-姿態分析................27
4.4.2 側拍畫面-速度分析................30
4.4.3 空拍畫面-頭部、肩膀軌跡..........30
第五章 結論與未來發展...................32
5.1 結論............................32
5.2 未來發展.........................32
參考文獻.................................34
Extended Abstract.......................36

[1]Zhao, Z. (2022). The Characteristics and Physical Fitness Characteristics of Track and Field Sprint Events Based on Discrete Dynamic Modeling of Complex Systems. Mobile Information Systems, 2022.
[2]Paradisis, G. P., & Cooke, C. B. (2001). Kinematic and postural characteristics of sprint running on sloping surfaces. Journal of Sports Sciences, 19(2), 149-159.
[3]陳德盛, 王月琪, 蔡宗晏, 張嘉澤, & 陳佳慧. (2016). 高中短距離選手 100 公尺速度, 反應與下肢力量分析. 臺灣體育學術研究, (60), 21-37.
[4]Hu, Y., & Alturjman, S. (2022, March). Timing System of Track and Field Competition Based on Data Analysis Algorithm. In The International Conference on Cyber Security Intelligence and Analytics (pp. 45-51). Springer, Cham.
[5]吳淑真. (2018). 臺灣體育運動科學現況與未來發展. 大專體育學刊, 20(3), i-vii.
[6]劉淑華, & 曾孝生. (2013). 200 公尺短跑選手爆發力與成績之相關研究. 運動教練科學, (29), 35-43.
[7]李灝 (2012) 國內100公尺短跑選手在高速期下肢擺動角度及速度與成績表現之研究. 2012. PhD Thesis. 競技運動學系碩士班.
[8]Romero-Franco, N., Martínez-López, E., Lomas-Vega, R., Hita-Contreras, F., & Martínez-Amat, A. (2012). Effects of proprioceptive training program on core stability and center of gravity control in sprinters. The Journal of Strength & Conditioning Research, 26(8), 2071-2077.
[9]曾景松, & 黃慶旻. (2009). 百米賽跑分段速度參數之相關探討. 大專體育, (100), 137-143.
[10]Villán, A. F. (2019). Mastering OpenCV 4 with Python: a practical guide covering topics from image processing, augmented reality to deep learning with OpenCV 4 and Python 3.7. Packt Publishing Ltd.
[11]Elgendy, N., & Elragal, A. (2014, July). Big data analytics: a literature review paper. In Industrial conference on data mining (pp. 214-227). Springer, cham.
[12]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).
[13]Redmon, J., & Farhadi, A. (2018). Yolov3: An incremental improvement. arXiv preprint arXiv:1804.02767.
[14]Albawi, S., Mohammed, T. A., & Al-Zawi, S. (2017, August). Understanding of a convolutional neural network. In 2017 international conference on engineering and technology (ICET) (pp. 1-6). Ieee.
[15]彭賢榮. (2019). 深度卷積神經網路中之自適應池化層設計.
[16]Neubeck, A., & Van Gool, L. (2006, August). Efficient non-maximum suppression. In 18th International Conference on Pattern Recognition (ICPR'06) (Vol. 3, pp. 850-855). IEEE.
[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.
[18]https://medium.com/daai/%E5%93%87-convolution-neural-network-%E5%8D%B7%E7%A9%8D%E7%A5%9E%E7%B6%93%E7%B6%B2%E7%B5%A1-%E9%80%99%E9%BA%BC%E7%89%B9%E5%88%A5-36d02ce8b5fe
[19]https://medium.com/%E9%9B%9E%E9%9B%9E%E8%88%87%E5%85%94%E5%85%94%E7%9A%84%E5%B7%A5%E7%A8%8B%E4%B8%96%E7%95%8C/%E6%A9%9F%E5%99%A8%E5%AD%B8%E7%BF%92-ml-note-convolution-neural-network-%E5%8D%B7%E7%A9%8D%E7%A5%9E%E7%B6%93%E7%B6%B2%E8%B7%AF-bfa8566744e9
[20]https://medium.com/jameslearningnote/%E8%B3%87%E6%96%99%E5%88%86%E6%9E%90-%E6%A9%9F%E5%99%A8%E5%AD%B8%E7%BF%92-%E7%AC%AC5-1%E8%AC%9B-%E5%8D%B7%E7%A9%8D%E7%A5%9E%E7%B6%93%E7%B6%B2%E7%B5%A1%E4%BB%8B%E7%B4%B9-convolutional-neural-network-4f8249d65d4f
[21]https://chih-sheng-huang821.medium.com/%E6%A9%9F%E5%99%A8-%E6%B7%B1%E5%BA%A6%E5%AD%B8%E7%BF%92-%E7%89%A9%E4%BB%B6%E5%81%B5%E6%B8%AC-non-maximum-suppression-nms-aa70c45adffa