羽球姿勢專家輔助系統之開發

Abstract

隨著科技日新月異，視覺辨識技術已被廣泛應用於運動產業中，若是想學羽球的初學者可能會需求專業教練的指導，依目前市場上羽球學費的行情，以台北為例，一對一教學的學費每小時400至上千元不等，場地費另計並且由學生負擔，對於初學者是很大的負擔及阻力，此外羽球揮拍的教學方式通常是教練一邊示範一邊口述講解，學生再模仿教練的動作來揮拍，教練在場邊適時指導及糾正動作，學生才能精準地完成揮拍動作，但僅僅透過課堂的訓練並不充足，課後也必須不斷的自主練習，因此，此研究研發一套羽球姿勢輔助教學系統給初學者使用，並給予羽球初學者以下的揮拍動作建議：1.持拍手手腕、手肘、肩膀之適當夾角。2.預備擊球時引拍手與持拍手之角度保持平衡之角度。3.揮拍時持拍手同側膝蓋需適當的彎曲之角度。4.揮拍次數計數功能。1至3項透過角度計算判斷預備擊球時引拍動作是否正確，以助於學習並且降低錯誤姿勢造成的運動傷害；第4項則是提供揮拍次數的資訊，資料部分收集共625張影像並進行Mediapipe座標軸分析，分析後的資料透過Logistic Regression、Random Forest Classifier、Support Vector Machine機器學習模型訓練並評比Accuracy、Recall rate，發現Random Forest Classifier 的Accuracy為97.0%，Recall rate則為97.3%表現最好，因此作為此系統計數模型。角度指標部分則是在欲計算夾角的揮拍前影像座標軸資料與相鄰兩點進行Atan2函式取得弧度，將弧度相減後取絕對值並轉換角度單位以計算手肘、肩膀、持拍手膝蓋彎曲夾角，再來透過紅綠燈角度指標系統整合各教練影像之揮拍角度資訊。最後透過Flask以及Bootstrap撰寫響應式全端網頁，並將模型以及紅綠燈角度系統整合，在台大體育館進行實測。實測人數5位並且皆無參加羽球社團或學球等經驗，結果部分顯示5位受試者平均擊長遠球之長度增加9.6%，擊球穩定度增加26.3%。總結以上，初學者在練習揮拍時將可以透過各尺寸聯網裝置使用此套系統，並有即時且正確的指引，不僅降低學球成本，對於提升球質亦十分有幫助，並同時降低受傷風險。

With the rapid advancement of technology, visual recognition techniques have been widely applied in the sports industry. However, for beginners who want to learn badminton, they may require the guidance of a professional coach. Considering the current market rates for badminton coaching fees, taking Taipei as an example, the tuition for one-on-one coaching can range from NT$400 to over a thousand per hour, excluding venue fees, which are paid by the students. This is a significant burden and obstacle for beginners. In addition, the teaching method for badminton swing typically involves the coach demonstrating and verbally explaining while the student imitates the coach's movements. The coach provides timely guidance and corrects the student's actions from the sidelines so that the student can accurately complete the swing. However, classroom training alone is not sufficient, and continuous self-practice is necessary after class. Therefore, this study developed a badminton posture-assisted teaching system for beginners. The system provides the following recommendations for beginners' swing movements: 1. Proper preparatory posture of the grip hand's wrist, elbow, and shoulder angle. 2. Balanced angle between the racket-hand and non-racket hand during the preparatory phase. 3. Appropriate angle of knee bending on the same side as the racket hand during the swing. 4. Swing count function. Item 1 to 3 aim to help in learning and also decrease the possibility of injuries by calculating the angles of the ready phase to determine whether the position is correct or not. A dataset of 625 images is collected and analyzed using the Mediapipe coordinate system. The analyzed data is then used to train and evaluate machine learning models, including Logistic Regression, Random Forest Classifier, and Support Vector Machine. Among these models, the Random Forest Classifier performed the best with an accuracy of 97% and a recall rate of 97.3%. Therefore, it is selected as the counting model for this system. Regarding the angle indicators, the Atan2 function is used to calculate the angle in radians from the image coordinate data of the previous frame and the two adjacent points before the swing. The absolute difference between the angles is then converted to degree units to calculate the angles of the elbow, shoulder, and knee bending. The red-green traffic light angle indicator system is integrated to combine the swing angle information from various coaching videos. Finally, a full-stack Responsive web page is developed by Flask and Bootstrap. The model and the red-green traffic light angle system are integrated into the webpage. The system is tested at the National Taiwan University Sports Center with a sample size of five participants, all of whom had no prior experience in joining badminton clubs or receiving formal badminton training. The result shows that the average length of hitting long shots increased by 9.6%, and the stability of hitting shots is improved by 26.3% among the five participants. In conclusion, beginners practicing their badminton swing can benefit from this system, which provides real-time and accurate guidance, not only reducing the cost of learning badminton but also significantly improving the quality of their shots and reducing the risk of injuries.