籃球投籃命中辨識暨自動化計分系統驗證

目的：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：錄製投籃影像，並將影像輸入以 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，共有五種不同目標投球數，每一個目標投球數皆有不同拍攝角度及不同背景環境的試投，共有十五個時間介於 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，並與人工紀錄進行比較。結果：對於籃球投籃出手的整體辨識準確率能達到 91.36%，對於籃球投籃進球的整體辨識準確率能達到 75.31%；在不同的拍攝角度與背景環境中，投籃出手及命中的辨識準確率皆無差異；在各個投籃位置中的出手及命中的辨識準確率也無差異，這表示此系統在室內籃球場中，不論在不同角度、背景環境及投籃位置都能夠進行穩定的辨識。結論：透過以機器學習為基礎的籃球投籃辨識系統，能夠紀錄投籃練習時的出手分布及投籃命中情形，雖然目前辨識效果有限，但未來此系統仍具有實行的可能性，往後可增加多人多球投籃影像，讓此系統持續學習、精進，也能更符合實際練習及比賽的的情境。

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

Purpose: The purpose of this study is to develop a basketball shooting recognition system. It is expected that this system will achieve rapid data analysis with simple equipment. Methods: The basketball shooting recognition system established with YOLOv4 was used to collect the data of actual shooting images. The images included four corners of the basketball half-court, and five different shooting target videos were recorded. Each target video had a different video shooting angle and background, with 15 videos ranging from 1 min 30 s to 5 min 30 s. The system was used to analyze the basketball shots and shooting goals, and the results were compared with manual records. Results: The overall recognition accuracy of basketball shots reached 91.36%, and the shooting goals reached 75.31%. In the different video shooting angles and backgrounds and different shooting positions on the court, there was no significant difference in the identification accuracy of basketball shots and shooting goals. Hence, the proposed system can be reliably identified in different video shooting angles, backgrounds, and basketball shooting positions on indoor basketball courts. Conclusion: Through the basketball shooting identification system based on machine learning, we can record the distribution of basketball shots and shooting goals. Although the identification effect is currently limited, it is still possible to implement this system in the future. In future studies, the images of multiplayer and multiball shooting should be increased so that the system can continue learning and improving to make it more consistent with actual practice and competitions.

並列關鍵字

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

參考文獻

Csapo, P., & Raab, M. (2014). "Hand down, man down." Analysis of defensive adjustments in response to the hot hand in basketball using novel defense metrics. PLoS One, 9(12), 1-25. https://doi.org/10.1371/journal.pone.0114184

Chen, H. T., Tien, M. C., Chen, Y. W., Tsai, W. J., & Lee, S. Y. (2009). Physics-based ball tracking and 3D trajectory reconstruction with applications to shooting location estimation in basketball video. Journal of Visual Communication and Image Representation, 20(3), 204-216. https://doi.org/10.1016/j.jvcir.2008.11.008

Chen, H. T., Chou, C. L., Fu, T. S., Lee, S. Y., & Lin , B-S. (2012). Recognizing tactic patterns in broadcast basketball video using player trajectory. Journal of Visual Communication and Image Representation, 23(6), 932-947. https://doi.org/10.1016/j.jvcir.2012.06.003

Vanhelst, J., Theunynck, D., Gottrand, F., & Béghin, L. (2010). Reliability of the RT3 accelerometer for measurement of physical activity in adolescents. Journal of Sports Sciences, 28(4), 375-379. https://doi.org/10.1080/02640410903502790

Wen, P. C., Cheng, W. C., Wang, Y-S., Chu, H. K., Tang, N. C., & Liao, H. Y. M. (2016). Court Reconstruction for Camera Calibration in Broadcast Basketball Videos. IEEE Transactions on Visualization and Computer Graphics, 22(5), 1517-1526. https://doi.org/10.1109/TVCG.2015.2440236

國際替代計量

籃球投籃命中辨識暨自動化計分系統驗證

全文下載

主題瀏覽