扯鈴為台灣民俗運動發展的重要項目,在國小亦是發展的重點項目。筆者為臺北市麗湖國小的扯鈴教練,深深覺得推廣扯鈴並不容易,培養一位扯鈴選手到一次可以運三顆扯鈴的程度,並能上場演出,需要至少3-4年以上的時間,而運三鈴並非人人可行,需要透過長時間的練習及適當的指導才能成功,希望能透過輔助的教學工具,增進教學品質,提升學會三鈴的成功率。 本研究將針對運三鈴的前置動作-三鈴抽線以機器學習的方式建置教學知識模型,透過收集大量的三鈴抽線影像,並以人工標記的方式分類專業及一般兩種等級,接著以機器學習的方式,以底下兩顆鈴的重心連線兩端之Y值差距7個像素以內的特徵找出關鍵畫面,再將連續關鍵畫面畫面以七種特徵的數值,作為分類模型的依據來分類專業組及一般組,而分類的準確率達89%,讓學習者能透過知識模型來自我擬定學習策略,進而提升學習效率,同時能以更精確的數值輔助教練判斷動作的好壞,並予以學員改進的回饋。
Diabolo is not only an important project for the development of Taiwan's folk sports, but also a key development project in elementary schools.The researcher is the coach of diabolo at Lihu Elementary School in Taipei, who deeply feel promotimg diabolo is hard, because training a diabolo player to play three diabolos at the same time and to be able to perform need at least 3-4 years, it takes a long time of practice and guidance to succeed. Through assisted teaching tools, the researcher hope teacher can improve teaching quality and increase the player’s success rate of playing three diabolos. The research used artificial intelligence to build a knowledge model of the stability of three diabolos hover start. By collecting a large number of images, and manual marking images to classify professional and general.Through machine learning, the threshold used to determine whether a frame is key frame or not was set to seven pixels to obtain all the key frames in the experiments. That is, frames having machine learning less or equal to seven had small enough y coordinate value differences. Then, seven image features extracted from the continuous key frames of each video were combined after feature elimination. The average and standard deviation groups had four and three relevant features, respectively. Using the seven features together achieved a accuracy 0f 89%. This research proposed using quantitative image features to extract key frames and to establish a classification model to separate the videos, hoping that the research can provide an automatic assessment to players without coaches.