2019 年新型冠狀病毒 (COVID-19) 爆發，教學型態因應環境因素改變，轉而邁向遠距教學，實體體育課程如何有效的轉型，成為此時空下值得深思的問題。動作捕捉（Motion capture）是項記錄並處理人或物體動作的技術，早期以穿戴裝置獲取數據，常被應用於遊戲動畫、電影製作、醫療、人因研究及體育等領域，人體姿態估計（Human Pose Estimation, HPE）為動作捕捉延伸技術，專門針對人體動作進行偵測，隨著深度學習引進，偵測人體關節數據僅需要影像即可達成。科技進步使人在面對問題時擁有更廣泛的選擇，應用上述技術於兩視頻人體動作比對，以數據方式客觀計算出動作間相似度，在實體體育課程轉型上，相信能提供一定的幫助。本研究建構出一個能普及化於應用的人體動作比對方法，以無標記動作捕捉方式，降低運動外在影響因素，所提出方法輸入端為深度學習人體姿態估計數據，經由正規化與數據延伸，可為不同型態的人物動作視頻進行比對。由於人體動作會受韻律影響不可能做到動作完全一致性，本研究利用動態時間扭曲演算法(Dynamic Time Warping, DTW) 找尋兩視頻最佳動作比對路徑，使動作比對更為彈性。本研究以多情境方式去驗證方法可靠性，針對動作完全相同、動作部分相同、動作完全不同情境進行結果檢測，實驗共 27 組比對視頻，動作比對相似度評分排序結果皆與實際相符，有此可證本研究所提出的動作比對方法具有一定的可行性。

With the outbreak of COVID-19 in 2019, the teaching mode had changed to distance learning due to the external environment. How to make physical education courses through distance education has become a problem that needs to be considered nowadays. Motion capture is a technique for recording the movement of people or objects. In the early days, it was necessary to obtain data through wearable sensing devices. Commonly used in animations, film productions, medical treatments, human factors research, sports, and other fields. Human pose estimation is an extension of motion capture technology. It is specially designed for human motion detection. With the introduction of deep learning, the detection of human joint data only needs to be done through videos. Advances in technology have given people a wider range of choices when facing problems. If the above-mentioned technology is used to compare the dynamic human movements in two videos, the similarity between the movements can be objectively calculated through the data and surely will improve the transformation of physical education. This research constructs a human body dynamic comparison method that can be generalized and applied. Adopt the unmarked dynamic capture method to reduce the external influence factors of the movement. The input of the method is deep learning human body pose estimation data, and the data is normalized and expanded. It can compare different types of dynamic videos of people. Because human movements are affected by rhythm, it is impossible to achieve the same movements. This study uses dynamic time warping to find the best dynamic comparison path between two videos, making the comparison method more accurate and flexible. This study uses a multi-case method to verify the method. Perform result detection when the actions are exactly the same, some of the actions are the same, and the actions are completely different. There are 27 sets of comparative videos in the experiment. The scoring results of the method are in line with expectations, which proves the feasibility of the method in practice.

中文摘要................................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...........................2
1.3 論文內容架構........................2
第二章 文獻探討..........................4
2.1 動作捕捉...........................4
2.2 人體姿態估計.......................7
2.2.1 問題與挑戰 ......................7
2.2.2 人體姿態估計方法 ................8
2.3 動作比對........................10
第三章 研究方法......................11
3.1 研究架構........................11
3.2 研究架構步驟說明..................13
3.2.1 人體動作數據獲取 ...............13
3.2.2 數據正規化 .....................14
3.2.3 關節數據延伸 ...................16
3.2.4 分數計算 .......................18
第四章 實驗設計與結果..................22
4.1 實驗說明..........................22
4.2 實驗情境設計.......................23
4.2.1 動作一：開合跳 ..................23
4.2.2 動作二：組合動作 ................23
4.2.3 動作三：八段錦 ..................24
4.3 人體動作比對實作...................27
4.4 數據結果分析......................32
4.5 後續動作改善建議..................34
第五章 結論與未來方向..................36
5.1 研究結論..........................36
5.2 未來研究方向......................37
參考文獻.............................38
Extended Abstract...................42

[1] 許珺佩、林耀豐（2010）。從定義與本質論點探討體育在教育上的價值。載於林瑞興主編，2010 年第三屆運動科學暨休閒遊憩管理學術研討會論文集(頁420-427)。屏東：國立屏東教育大學。
[2] Cao, Z., Hidalgo, G., Simon, T., Wei, S. E., & Sheikh, Y. (2019). OpenPose: realtime multi-person 2D pose estimation using Part Affinity Fields. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(1), 172-186
[3] Fang, H. S., Xie, S., Tai, Y. W., & Lu, C. (2017). Rmpe: Regional multi-person pose estimation. Proceedings of the IEEE International Conference on Computer Vision,2334-2343.
[4] Güler, R. A., Neverova, N., & Kokkinos, I. (2018). Densepose: Dense human pose estimation in the wild. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 7297-7306.
[5] Oved, D., Pisapia, A., & Gudnason, A. (2020, June 16). Estimating 3D Poses of Athletes at Live Sporting Events. The New York Times, form the World Wide Web:
https://rd.nytimes.com/projects/estimating-3d-poses-of-athletes-at-live-sportingevents
[6] Nogueira, P. (2011). Motion capture fundamentals – a critical and comparative analysis on real world applications. Proceedings of the 7th Doctoral Symposium in Informatics Engineering, 303.
[7] Grant, F. S., Guilford, J. P., & West, G. F. (1966). Interpretation Theory in Applied Geophysics. New York: McGraw-Hill, 306-381.
[8] Marshall, S. V. (1978). Vehicle detection using a magnetic field sensor. IEEE Transactions on Vehicular Technology, 27(2), 65-68.
[9] Wynn, W., Frahm, C., Carroll, P., Clark, R., Wellhoner, J., & Wynn, M. (1975). Advanced superconducting gradiometer/magnetometer arrays and a novel signal processing technique. IEEE Transactions on Magnetics, 11(2), 701-707.
[10] Hashi, S., Tokunaga, Y., Yabukami, S., Toyoda, M., Ishiyama, K., Okazaki, Y., & Arai, K. I. (2005). Development of real-time and highly accurate wireless motion capture system utilizing soft magnetic core. IEEE Transactions on Magnetics, 41(10), 4191-4193.
[11] Ligorio, G., & Sabatini, A. M. (2016). Dealing with magnetic disturbances in human motion capture: A survey of techniques. Micromachines, 7(3), 43.
[12] van der Kruk, E., & Reijne, M. M. (2018). Accuracy of human motion capture systems for sport applications; state-of-the-art review. European Journal of Sport Science, 18(6), 806-819.
[13] Wu, T., McGinley, J., Duffy, V. G., & Liu, L. (2005). Application and validation of a mechanical motion capture-based industrial ergonomics assessment system (No. 2005-01-2733). SAE Technical Paper.
[14] Guerra-Filho, G. (2005). Optical Motion Capture: Theory and Implementation. RITA, 12(2), 61-90.
[15] Hurwitz, A. M., & Dotterweich, J. (2019). Vicon System Precision Analysis at US Army Combat Capabilities Development Command Army Research Laboratory. Combat Capabilities Development Command Army Research Laboratory Aberdeen Proving Ground United States.
[16] Vlasic, D., Adelsberger, R., Vannucci, G., Barnwell, J., Gross, M., Matusik, W., & Popović, J. (2007). Practical motion capture in everyday surroundings. ACM Transactions on Graphics (TOG), 26(3), 35-es.
[17] Jang, M., Kim, J. S., Kang, K., Um, S. H., Yang, S., & Kim, J. (2019, March). Development of Wearable Motion Capture System Using Fiber Bragg Grating Sensors for Measuring Arm Motion. In 2019 IEEE Conference on Virtual Reality and 3D User Interfaces (VR) (pp. 994-995). IEEE.
[18] Colyer, S. L., Evans, M., Cosker, D. P., & Salo, A. I. (2018). A review of the evolution of vision-based motion analysis and the integration of advanced computer vision methods towards developing a markerless system. Sports Medicine-
Open, 4(1), 1-15.
[19] Smisek, J., Jancosek, M., & Pajdla, T. (2011, November). 3D with Kinect. In 2011 IEEE International Conference on Computer Vision Workshops (ICCV Workshops) (pp. 1154-1160). IEEE.
[20] Agarwal, A. (2006). Machine learning for image based motion capture (Doctoral dissertation, Institut National Polytechnique de Grenoble-INPG).
[21] Li, B., Chen, H., Chen, Y., Dai, Y., & He, M. (2017, July). Skeleton boxes: Solving skeleton based action detection with a single deep convolutional neural network. In 2017 IEEE International Conference on Multimedia & Expo Workshops
(ICMEW) (pp. 613-616). IEEE.
[22] Li, B., Dai, Y., Cheng, X., Chen, H., Lin, Y., & He, M. (2017, July). Skeleton based action recognition using translation-scale invariant image mapping and multi-scale deep CNN. In 2017 IEEE International Conference on Multimedia & Expo Workshops (ICMEW) (pp. 601-604). IEEE.
[23] Luvizon, D. C., Picard, D., & Tabia, H. (2018). 2d/3d pose estimation and action recognition using multitask deep learning. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 5137-5146.
[24] Li, B., He, M., Dai, Y., Cheng, X., & Chen, Y. (2018). 3D skeleton based action recognition by video-domain translation-scale invariant mapping and multi-scale dilated CNN. Multimedia Tools and Applications, 77(17), 22901-22921.
[25] Insafutdinov, E., Andriluka, M., Pishchulin, L., Tang, S., Levinkov, E., Andres, B., & Schiele, B. (2017). Arttrack: Articulated multi-person tracking in the wild. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 6457-6465.
[26] Rogez, G., Rihan, J., Ramalingam, S., Orrite, C., & Torr, P. H. (2008, June). Randomized trees for human pose detection. In 2008 IEEE Conference on Computer Vision and Pattern Recognition (pp. 1-8). IEEE.
[27] Urtasun, R., & Darrell, T. (2008, June). Sparse probabilistic regression for activityindependent human pose inference. In 2008 IEEE Conference on Computer Vision and Pattern Recognition (pp. 1-8). IEEE.
[28] Pishchulin, L., Andriluka, M., Gehler, P., & Schiele, B. (2013). Strong appearance and expressive spatial models for human pose estimation. Proceedings of the IEEE International Conference on Computer Vision, 3487-3494.
[29] Andriluka, M., Roth, S., & Schiele, B. (2009, June). Pictorial structures revisited: People detection and articulated pose estimation. In 2009 IEEE Conference on Computer Vision and Pattern Recognition (pp. 1014-1021). IEEE.
[30] Ionescu, C., Li, F., & Sminchisescu, C. (2011, November). Latent structured models for human pose estimation. In 2011 International Conference on Computer Vision (pp. 2220-2227). IEEE.
[31] Pishchulin, L., Andriluka, M., Gehler, P., & Schiele, B. (2013). Poselet conditioned pictorial structures. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 588-595.
[32] Toshev, A., & Szegedy, C. (2014). Deeppose: Human pose estimation via deep neural networks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 1653-1660.
[33] Andriluka, M., Pishchulin, L., Gehler, P., & Schiele, B. (2014). 2d human pose estimation: New benchmark and state of the art analysis. Proceedings of the IEEE Conference on computer Vision and Pattern Recognition, 3686-3693.
[34] Gkioxari, G., Hariharan, B., Girshick, R., & Malik, J. (2014). Using k-poselets for detecting people and localizing their keypoints. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 3582-3589.
[35] Dantone, M., Gall, J., Leistner, C., & Van Gool, L. (2013). Human pose estimation using body parts dependent joint regressors. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 3041-3048.
[36] 馮曉月與宋杰（2020）。二維人體姿態估計研究進展。計算機科學，47(11)，128-136。
[37] Fischler, M. A., & Elschlager, R. A. (1973). The representation and matching of pictorial structures. IEEE Transactions on Computers, 100(1), 67-92.
[38] Andriluka, M., Roth, S., & Schiele, B. (2009, June). Pictorial structures revisited:People detection and articulated pose estimation. In 2009 IEEE Conference on Computer Vision and Pattern Recognition (pp. 1014-1021). IEEE.
[39] Andriluka, M., Pishchulin, L., Gehler, P., & Schiele, B. (2014). 2d human pose estimation: New benchmark and state of the art analysis. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 3686-3693.
[40] Cao, Z., Simon, T., Wei, S. E., & Sheikh, Y. (2017). Realtime multi-person 2d pose estimation using part affinity fields. Proceedings of the IEEE conference on computer vision and pattern recognition, 7291-7299.
[41] Sun, K., Xiao, B., Liu, D., & Wang, J. (2019). Deep high-resolution representation learning for human pose estimation. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 5693-5703.
[42] 圖像算法（2019 年 8 月 24 日）：單人或多人的人體姿態骨架估計算法彙總。PCNOW。網址：https://pcnow.cc/p/OdY9gd6daa.html/
[43] Iqbal, M., Suhendra, A., & Mutiara, A. B. (2018, October). Similarity Analysis of Taekwondo Movement Using Data Motion. In 2018 Third International Conference on Informatics and Computing (ICIC) (pp. 1-6). IEEE.
[44] Ota, M., Tateuchi, H., Hashiguchi, T., Kato, T., Ogino, Y.,Yamagata, M., & Ichihashi, N. (2020). Verification of reliability and validity of motion analysis systems during bilateral squat using human pose tracking algorithm. Gait &
posture, 80, 62-67.
[45] Gámez Díaz, R. (2021). Digital Twin Coaching for Edge Computing Using Deep Learning Based 2D Pose Estimation (Doctoral dissertation, Universitéd'Ottawa/University of Ottawa).
[46] Ruan, X., & Tian, C. (2015, August). Dynamic gesture recognition based on improved DTW algorithm. In 2015 IEEE International Conference on Mechatronics and Automation (ICMA) (pp. 2134-2138). IEEE.
[47] 吳貴崗（2011）。使用 Kinect 體感攝影機藉由人體骨架進行人類動作識別。（碩士論文，國立台北科技大學，2011）。國立臺北科技大學資訊工程系研究所碩士論文，1-65。
[48] Yasin, H., & Hayat, S. (2021). DeepSegment: Segmentation of motion capture data using deep convolutional neural network. Image and Vision Computing, 109, 104147.