透過您的圖書館登入
IP:18.116.40.47
  • 學位論文

基於人體姿勢估計之跆拳道側踢分析

Quality Assessment of Taekwondo Side Kick Stance Based on Human Pose Estimation

指導教授 : 賀耀華
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


在跆拳道品勢 (Taekwondo Poomsae) 比賽或訓練中,評估該運動表現唯基於專家及教練的觀察,並根據其自身經驗會有不同的想法,存在多種公平性問題,此外,教練也無法全天候指導所有學員,人們對於量化評價方法和工具之需求日益增加。然而,跆拳道快速的肢體動作與結構極端繁複的技術,使量化困難且不易評估。 跆拳道品勢單元技術中,側踢 (Side Kick) 屬於較複雜、評分比重較高的項目,因此,本論文針對側踢先行試驗,我們以臺北市立龍山國中以及國立臺灣師範大學的跆拳道品勢選手作為研究對象,並參考專家建議之評分標準,提出基於人體姿勢估計 (Human Pose Estimation) 之跆拳道側踢分析,通過專業認證的跆拳道側踢評分系統 (Taekwondo Side Kick Assessment System, SideKick),能夠有效地量化選手運動數據,分析並評估其側踢表現。 本研究中,我們首先建立了具高度公信力的跆拳道側踢資料集,由專業品勢教練進行動作質量評分;接著透過人體姿勢估計的方式,偵測人體關節點座標,精確獲取肢體運動角度及高度變化數據,使得運動特徵不易受場景影響,將攝影鏡頭校正難度降低;最後,我們參考專家提供之側踢建議量化特徵,分析各特徵的重要性排序,並利用機器學習的方式,訓練運動時空特徵及專家建議特徵,來預測選手整體側踢表現分數。 實驗以均方根誤差與交叉驗證評估多種回歸模型方法,最終選擇卷積神經網路模型,作為系統之評分模組。結果顯示實際應用之誤差為0.69,經信度檢驗,其結果也達顯著相關,在容許誤差為1的範圍內,準確率達86%。本研究提出之SideKick系統不需花費大量金錢及人力,且錄製設備取得容易。學員們能藉由本系統了解自身能力,教練們也可以在不限任何時間或地點下指導學員,提升團體訓練效益,並為未來遠程跆拳道品勢評價系統奠定基礎。

並列摘要


In Taekwondo Poomsae training and competition, the only way to evaluate performance is based on the observations of experts and coaches. On account of their different experience and possibly biased appraisals, there is a growing demand for quantitative assessment methods and tools. However, the rapid physical movements and extremely complex structure of Taekwondo techniques make it difficult to quantify and evaluate. The side kick is one of the more complex and heavily weighted aspects of the Taekwondo technique. Therefore, this paper uses side kick as a first test, we propose a Taekwondo side kick assessment system based on Human Pose Estimation called SideKick. Our SideKick system enables us to quantify the athletes’ side kick data, then it can automatically analyze and evaluate their performance. In the proposed SideKick system in this study, we first established a highly credible Taekwondo side kick dataset scored by professional judges and coaches. Then, we used human pose estimation to detect the coordinates of the body's joints and obtain precise data on the angle and height changes of the body movement. Finally, we analyze these values relying on the experts’ suggestions and use machine learning to assess the quality of side kick.

參考文獻


賀敏彥、李佳融、張家豪 (2020)。中國跆拳道品勢執裁標準現狀研究。跆拳道學刊,7 期,83- 91 頁。
品勢「跆」雙銅 總教練:裁判太主觀. Retrieved June 10, 2022, from https://sports.ltn.com.tw/news/paper/1225909
Ahmadi, A., Rowlands, D., & James, D. A. (2009). Towards a wearable device for skill assessment and skill acquisition of a tennis player during the first serve. Sports Technology, 2(3-4), 129-136.
Pueo, B. (2016). High speed cameras for motion analysis in sports science. Journal of Human Sport and Exercise, 11(1), 53-73.
Tran, D., Ray, J., Shou, Z., Chang, S.-F., & Paluri, M. (2017). Convnet architecture search for spatiotemporal feature learning. ArXiv Preprint ArXiv:1708.05038.

延伸閱讀