台灣羽球選手成功站上世界舞台發光發熱，屢次獲得優異成績。機器視覺、深度學習融合人體關鍵點之羽球動作分類辨識模型，使用更科學的方法輔助羽球教練與選手，讓選手不會受到穿戴設備影響，讓教練可以在賽前、賽後使用選手與對手的影片，更簡易與自動化進行動作統計分析。 本論文的方法為連續擷取影像中人體骨架關鍵點座標，錄製10位羽球選手之13類正反手羽球動作影片，將每段影片8:2分割建立彼此獨立之訓練與測試數據集（dataset）共計訓練182,780筆及測試1,300筆資料，訓練資料型態為四維矩陣，每筆訓練資料為具有上下文關聯36個Frames的羽球動作，採用深度學習CNN神經網路建模，訓練模型辨識羽球動作。實驗分兩階段進行，第一階段為單人模型，先使用相同模型架構與參數比較3種攝影角度之辨識準確率，再進行9位選手單人模型羽球動作分類辨識實測，單人模型平均準確率81.28%，第二階段為多人模型，1號~8號選手資料為訓練集，9號選手資料為測試集，準確率為55.38%。 本論文動作辨識模型訓練所需的數據，使用自行建立的人體骨架關鍵點座標數據集，而不是常見的影片或圖像，這具有去隱私化的優點，後續相關的研究將不再受限於個資與肖像權保護，而關鍵點座標數據集相較於影像辨識，運算資料量大幅減少，有利於邊緣運算與低成本優勢，也不限於羽球單項運動，可推廣至其他體育運動。

Taiwanese badminton players have successfully shown up on international tournaments and repeatedly conducted the outstanding performance. Computer vision and deep learning let the coaches, and players can easily analyze the recorded video of the games. Besides, comparing to traditional methods, players will not be affected by wearable devices during the training. This will greatly improve the benefit and accuracy of the badminton movement images recognition.
The main point of the method used by this thesis is to extract the coordinates of key points of the human skeleton from images. Those images include 13 types of badminton action data collected from 10 badminton players. The training data has 182,780(80%) piece of data. On the other hand, the testing data has 1,300(20%) piece of data. Besides, the training data type is a four-dimensional matrix and each data includes 36 frames of badminton movements. The model uses deep learning CNN neural network to identify badminton movements. The experiment was conducted in two stages. The first stage is the personal badminton movement recognition model by using the badminton movement dataset established in this thesis. First of all, this research compared the accuracy of three cameras with different shooting angles and the same parameters. Next, it conducted action recognition and classify from 9 players. In this model, the accuracy is 81.28%. The second stage is the multiplayer badminton movement recognition model by using the data of player numbers 1 to 8 to train the model and the data of player number 9 for testing. The accuracy of the second model is 55.38%.
In this thesis, the data required for the training of the action recognition model used the self-built human skeleton key point coordinate dataset instead of the commonly used movies or images. This brings the advantage of deprivation and subsequent related research will no longer be limited to the protection of data and portrait rights. Moreover, the key point of the coordinate dataset is that it can greatly reduce the computing data comparing to image recognition. This significant benefit is conducive to edge computing and can lower the cost. Furthermore, the method used in this thesis is not limited to badminton, but can widely extend to other sports. This will facilitate sports training resources and sports posture recognition research.

[1] 陳奕仁, 方介民, 陳金海, 張世沛, 張富貴,“我國羽球運動相關研究趨勢之探討,”長榮大學, 2015.
[2] 蔡忠信,“我國羽球運動推展職業化之可行性-以羽球從業人員及選手為例,“國立台東大學,2007.
[3] 曾元正, “臺灣羽球運動員企業贊助之研究,” 國立臺灣師範大學, 2019.
[4] 鄭健民, “基層羽球運動教練專業能力現況與需求之研究,”國立台東大學,2008.
[5] 姚漢禱, 陳永銘, 王榮照, “羽球單打比賽技術診斷理論與實務分析,”, 國立體育大學, 2014.
[6] The Badminton World Federation (BWF), “The BWF Statutes LAWS OF BADMINTON”, The Badminton World Federation (BWF), 2020.
[7] 陳毅桓. “戰術?! 戰略?! 化繁為簡-量化羽球單打戰術分析, ”,國立臺灣師範大學, 2012.
[8] S. Ramasinghe, K. G. M. Chathuramali, R. Rodrigo, “Recognition of badminton strokes using dense trajectories,” University of Moratuwa Sri Lanka, 2014
[9] W. T. Chu, S. Situmeang , “Badminton video analysis based on spatiotemporal and stroke features,”ICMR, 2017, pp. 448-451.
[10] M. Promit, S. Waijanya, “ Model for Practice Badminton Basic Skills by using Motion Posture Detection from Video Posture Embedding and One-Shot Learning Technique”, In: Proceedings of the 2019 2nd Artificial Intelligence and Cloud Computing Conference. 2019, pp.117-124.
[11] Y. Wang, M. Chen, X. Wang, R. H. M. Chan, W. J. Li, “IoT for next-generation racket sports training”, IEEE Internet of Things Journal, 2018, vol 5 Dec-2018, pp. 4558 - 4566.
[12] Z. Wang, M. Guo, C. Zhao, “Badminton stroke recognition based on body sensor networks”, IEEE Transactions on Human-Machine Systems, 2016, vol 46, pp. 769 - 775.
[13] Y. Zhu, F. Kikuo, “A bayesian framework for human body pose tracking from depth image sequences,” Sensors 10.5 , May-2010, 2010, pp. 5280-5293.
[14] G. Mori, M. Jitendra, “Recovering 3d human body configurations using shape contexts”, IEEE Transactions on Pattern Analysis & Machine Intelligence 7:, Jul 2006, 2006, vol28, pp. 1052-1062.
[15] Z. Cao, T. Simon, S.-E. Wei, Y. Sheikh, “Realtime multi-person 2d pose estimation using part affinity fields”, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition., Nov 2016.
[16] D. G. Lowe,“Distinctive image features from scale-invariant keypoints,”International journal of computer vision 60.2: , Jan 2004, pp. 91-110.
[17] A. Collet, M. Martinez, S. S Srinivasa, “The MOPED framework: Object recognition and pose estimation for manipulation”, The International Journal of Robotics Research 30.10. , Apr 2011, pp. 1284-1306.
[18] S. Nowozin, J. Shotton, “Action points: A representation for low-latency online human action recognition”, Microsoft Research Cambridge, Tech. Rep. MSR-TR-2012-68., Jul 2012.
[19] A. Nibali, Z. He, S. Morgan, L. Prendergast, “ 3d human pose estimation with 2d marginal heatmaps”, In: 2019 IEEE Winter Conference on Applications of Computer Vision (WACV). IEEE, 2019.
[20] H.-S. Fang, S. Xie, Y.-W. Tai, C. Lu, “Rmpe: Regional multi-person pose estimation,” Proceedings of the IEEE International Conference on Computer Vision., Dec 2016.
[21] R. A. Güler, N. Neverova, I. Kokkinos “Densepose: Dense human pose estimation in the wild”, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Feb 2018.
[22] F. C. Heilbron, V. Escorcia, B. Ghanem, J. C. Niebles, “Activitynet: A large-scale video benchmark for human activity understanding”, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition., Jun 2015.
[23] E. P. Volkova, B. J. Mohler, T. J. Dodds, J. Tesch, H. H. Bülthoff,“Emotion categorization of body expressions in narrative scenarios”, Frontiers in psychology, 2014.