應用人工智慧及機器學習進行籃球投籃命中辨識

緒論：大數據及人工智慧結合籃球影像數據分析的研究已蔚為風潮，反觀國內在數據分析上的應用，多數仍以人工記錄的方式，耗費人力與時間，若能透過機器學習將使數據分析記錄更為快速。因此本研究目的為建構籃球投籃辨識系統，期望運用此系統能以簡易器材達成快速的數據分析，增進球隊訓練及比賽的效率。方法：以YOLOV4所建構的籃球投籃辨識系統，進行實際投籃影像的分析，拍攝畫面包含籃球半場的角落四點，進行六種不同目標投球數，每一個目標投球數皆有兩名實驗參與者，共有十二個時間介於1分30秒到5分30秒的影像片段，讓此系統進行投籃出手及投籃進球的判定，並與人工記錄進行比較。結果：對於籃球投籃出手的整體辨識準確率能達到94%，對於籃球投籃進球的整體辨識準確率能達到81%，並以視覺化的圖表呈現。結論：透過以機器學習為基礎的籃球投籃辨識系統，在不受設備資源及人力資源的限制下，能夠記錄投籃練習時的出手分布及投籃命中情形，並以視覺化圖表呈現。

關鍵字

機器視覺；籃球自主訓練系統；投籃熱區；投籃命中率

並列摘要

Introduction: The research of big data and artificial intelligence combined with basketball image data analysis has become a trend. Looking at the application of data analysis in Taiwan, most of them are still recorded manually, which consumes manpower and time. If machine learning can be used, data analysis and recording will be faster. Therefore, the purpose of this study is to develop the Basketball Shooting Recognition System. It is expected that this system can achieve rapid data analysis with simple equipment and improve the efficiency of team training and competition. Methods: The basketball shooting recognition system constructed by YOLOV4 was used to collect the data of actual shooting images. The images included four corners of the basketball half court. And six different shooting target videos were carried out. Each target had two experimental participants, with a total of 12 video ranging from 1 minute 30 seconds to 5 minute 30 seconds. Let the system judge the shooting and shooting goal, and compare it with the manual record. Results: The overall recognition accuracy of basketball shooting shot can reach 94%. And the overall recognition accuracy of basketball shooting goal can reach 81%. All of them can be presented in visual charts. Conclusion: Through the basketball shooting recognition system based on machine learning. Without the limitation of equipment resources and human resources. It can record the shooting distribution and shooting goal during shooting practice, and present them in visual charts. However, due to the current technological development, the recognition results are still vulnerable to the background environment. So, accumulate enough data, making machines learn continuously is the direction of efforts in the future.

並列關鍵字

machine vision ； basketball self-training system ； shot chart ； field goal percentage

參考文獻

李逸驊、蔡琪揚、陳韋翰、黃冠勛、戴沁琳 (2019)。穿戴加速規測量大專籃球聯賽公開男生組第一級隊伍之運動強度。大專體育學刊，21(4)，342-352。https://doi.org/10.5297/ser.201912_21(4).0004

李雲光 (2007)。不同投籃姿勢各關節運動學分析。大專體育學刊，9(4)，113-123。https://doi.org/10.5297/ser.200712_9(4).0009

李建興、游凱倫、林應璞 (2010)。即時動態車牌辨識。技術學刊，25(2)，151-165。https://doi.org/10.29507/JT.201006.0007

周育晨、李恆儒 (2020)。以穿戴式裝置探討不同專項位置籃球員與訓練情境之運動負荷。體育學報，53(3)，315-326。https://doi.org/10.6222/pej.202009_53(3).0004

相子元、石又、何金山 (2012)。感測科技於運動健康科學之應用。體育學報，45(1)，1-12。https://doi.org/10.6222/pej.4501.201203.0701

主題瀏覽