影像動作辨識系統應用於跆拳道競技與品勢動作分析的效益：系統性回顧

Effect of video action recognition system on Taekwondo competition and Poomsae motion analysis: A systematic review

跆拳道競技與品勢在傳統情報蒐集和教學上，有著需耗費大量時間與教學上過於主觀的限制。本研究透過系統性回顧彙整2017年4月至2024年4月，共7年間，影像動作辨識系統應用於跆拳道競技與品勢的相關實證研究，釐清適用何種深度學習模式、基於哪種提取法、參考哪些數據資料庫、適用哪種層級選手、能分析多少影片事件數量、辨識召回率、準確率與識別率為何，與具體能觀測哪些事件（動作技術）；並提出實務應用、未來研究方向與相關建議，供跆拳道教練團與情蒐人員深入瞭解影像動作辨識系統如何輔助競賽情蒐與提升品勢教學效益。

Traditional information gathering and teaching in Taekwondo competition and Poomsae had limitations, requiring a significant amount of time and being overly subjective. This study systematically reviewed and compiled empirical research related to the application of image motion recognition systems in Taekwondo competition and Poomsae from April 2017 to April 2024, a total of 7 years. It clarified which deep learning models were applicable, which extraction methods were used, which data databases were referenced, the levels of players to which it was applicable, the number of video events that could be analyzed, the recall rate, accuracy, and identification rate, and specifically which events (motion techniques) could be observed. Practical applications, future research directions, and related recommendations were proposed for Taekwondo coaching teams and information gathering personnel to deeply understand how image motion recognition systems assist in competition information gathering and improve the effectiveness of Poomsae teaching.