本研究提出一套整合電腦視覺落點偵測與混合二階馬可夫模型的低成本半自動化技戰術分析框架，並以中國頂尖女子桌球選手孫穎莎的比賽資料進行實證分析。此框架旨在降低傳統人工編碼所需的時間與人力成本，改善長時間編碼過程中一致性難以維持的問題，並透過技戰術轉移關係的模型建構與打法調整模擬，提供較具體且直觀的戰術調整依據。
在電腦視覺方面，本研究以非固定機位的公開轉播影像為資料來源，採用兩階段訓練策略，分別訓練創新的雙分類頭BL-YOLOv8s物件偵測模型與YOLOv8s-pose姿態估計模型。雙分類頭可同時偵測ball（B）與landing point（L），使落點判定同時結合球體軌跡幾何資訊與落點直接偵測結果，並搭配逐幀桌面定位與霍夫轉換角點校正機制，以因應攝影機平移、縮放及視角變化所造成的桌面角點漂移，提升俯瞰座標轉換的幾何穩定性。在橫向視角影像中，系統的寬容準確率達94.6%，接近人工編碼者間一致性的參考標準；在斜向與縱向視角影像中，亦達到63.2%的落點偵測率，顯示系統具備處理不同轉播視角與非固定機位影像的能力。
在技戰術分析方面，依孫穎莎和對手的發球與接發球情境建構四種混合二階馬可夫鏈模型（M1，M2，M3，M4）。本研究創新之處更在於提出以群組屬性為單位的打法調整模擬方法，將擾動對象從單一打法狀態擴展至具有相同屬性的打法群組。分析結果顯示，孫穎莎無論在發球後第三板或接發球後第四板，若能快速銜接正手進攻並將球落至正手長球位，勝率均有顯著提升；相反地，若進入反手控制或防禦型銜接，勝率則出現明顯下降。
本研究所建立之分析框架，驗證了以低成本公開影像半自動完成落點辨識的可行性，並提供一套從序列轉移視角出發、具備直觀解釋力的技戰術優化工具，可進一步應用於其他選手或球類運動之技戰術分析。

This study proposes a low-cost, semi-automated technical and tactical analysis framework that integrates computer vision-based landing point detection with a mixture second-order Markov model. Match data from Sun Yingsha, an elite Chinese female table tennis player, were used for empirical analysis. The framework aims to reduce the time and labor costs required for conventional manual coding, address the difficulty of maintaining coding consistency during prolonged coding sessions, and provide more specific and intuitive guidance for tactical adjustment through the modeling of technical-tactical transition relationships and play adjustment simulations.
For the computer vision component, publicly available broadcast footage captured by non-fixed cameras was used as the data source. A two-stage training strategy was adopted to train an innovative dual-classification-head BL-YOLOv8s object detection model and a YOLOv8s-pose pose estimation model. The dual-classification-head design simultaneously detects the ball (B) and landing point (L), allowing landing point determination to combine geometric information derived from the ball trajectory (B) with direct landing point detection results (L). A Hough Transform-based corner refinement mechanism was further applied to suppress table-corner drift caused by camera translation, zooming, and viewpoint changes, thereby improving the geometric stability of the homography transformation to overhead coordinates. For lateral-view images, the system achieved a relaxed accuracy of 94.6% for landing point detection, approaching the reference standard for inter-coder agreement in manual coding. For oblique-view and longitudinal-view images, the system achieved a landing point detection rate of 63.2%, demonstrating its ability to process non-fixed-camera footage across different broadcast views.
For the technical and tactical analysis, four mixture second-order Markov models, denoted as M1, M2, M3, and M4, were constructed according to the serving and receiving scenarios of Sun Yingsha and her opponents. A further contribution of this study is the development of a grouped play adjustment simulation method, which extends the perturbation target from an individual play state to groups of plays sharing the same attribute. The results showed that Sun Yingsha’s winning probability increased substantially when she quickly transitioned to a forehand attack and directed the ball to the deep forehand landing zone, either on the third stroke after serving or on the fourth stroke after receiving serve. In contrast, transitions involving backhand control or defensive plays resulted in noticeable decreases in winning probability.
The proposed framework demonstrates the feasibility of using low-cost, publicly available video footage for semi-automated landing point detection. It also provides an interpretable technical and tactical optimization tool based on sequential transition relationships, and can be further extended to the analysis of other players and other ball sports.

[1] Zilin He, Zeyi Yang, Jiarui Xu, Hongyu Chen, Xuanfeng Li, Anzhe Wang, Jiayi Yang, Gary Chi-Ching Chow, and Xihan Chen. Real-time accurate determination of table tennis ball and evaluation of player stroke effectiveness with computer vision-based deep learning. Applied Sciences, 15(10):5370, 2025.
[2] Ross Girshick, Jeff Donahue, Trevor Darrell, and Jitendra Malik. Rich feature hierarchies for accurate object detection and semantic segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 580–587, 2014.
[3] Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 779–788, 2016.
[4] Cheng-Wen Wu, Ming-Der Shieh, Jenn-Jier James Lien, Jar-Ferr Yang, Wei-Ta Chu, Tsang-Hai Huang, Han-Chuan Hsieh, Hung-Ta Chiu, Kuo-Cheng Tu, Yen-Ting Chen, Shian-Yu Lin, Jia-Jun Hu, Chen-Huan Lin, and Cheng-Siang Jheng. Enhancing fan engagement in a 5G stadium with AI-based technologies and live streaming. IEEE Systems Journal, 16(4):6590–6601, 2022.
[5] Guan-Yu Wu, Chun-Ho Hung, Hsuan-Wei Chen, and Wei-Ta Chu. A trajectory-based statistics and tactics analysis system for table tennis. In Proceedings of ACM Multimedia Asia 2023 (MMAsia), pages 1–3, 2023.
[6] Tao Ning, Meng Fu, Yuzhe Wang, Xiaodong Duan, and Mohammad Zoynul Abedin. Application of deep learning for automatic detection of table tennis balls from an intelligent serving machine. Applied Soft Computing, 167:112329, 2024.
[7] Ze-Hong Lan. Study on ping pong match landing point detection and score judgment based on YOLOv8. Master’s thesis, Department of Telecommunication Engineering, National Kaohsiung University of Science and Technology, Kaohsiung, Taiwan, 2025.
[8] Richard O. Duda and Peter E. Hart. Use of the Hough transformation to detect lines and curves in pictures. Communications of the ACM, 15(1):11–15, 1972.
[9] Thomas Gossard, Andreas Ziegler, and Andreas Zell. TT3D: Table tennis 3d reconstruction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 5867–5877, 2025.
[10] Chung-Yi Wang Li and Che-Wei Chuang. Comparison of two different analysis methods’ impact on table tennis technical performance recording. Journal of Sports Research, 33(2):109–125, 2024.
[11] Honglin Song, Yutao Li, Chenyi Fu, Feng Xue, Qiyue Zhao, Xingyu Zheng, Kunkun Jiang, and Tianbiao Liu. Using complex networks and multiple artificial intelligence algorithms for table tennis match action recognition and technical-tactical analysis. Chaos, Solitons & Fractals, 178:114343, 2024.
[12] Mark Pfeiffer, Hui Zhang, and Andreas Hohmann. A Markov chain model of elite table tennis competition. International Journal of Sports Science & Coaching, 5(2):205–222, 2010.
[13] Hui Zhang, Zheng Zhou, and Qing Yang. Match analyses of table tennis in china: A systematic review. Journal of Sports Sciences, 36(23):2663–2674, 2018.
[14] Javier Galeano, Miguel-Ángel Gómez, Fernando Rivas, and Javier M. Buldú. Using Markov chains to identify player’s performance in badminton. Chaos, Solitons & Fractals, 165:112828, 2022.
[15] Jiachen Wang, Kejian Zhao, Dazhen Deng, Anqi Cao, Xiao Xie, Zheng Zhou, Hui Zhang, and Yingcai Wu. Tac-Simur: Tactic-based simulative visual analytics of table tennis. IEEE Transactions on Visualization and Computer Graphics, 26(1):407–417, 2020.
[16] Sida Peng, Eirini Kalliamvakou, Peter Cihon, and Mert Demirer. The impact of AI on developer productivity: Evidence from GitHub Copilot. arXiv preprint arXiv:2302.06590, 2023.
[17] Glenn Jocher, Ayush Chaurasia, and Jing Qiu. Ultralytics YOLOv8. Ultralytics Software, https://github.com/ultralytics/ultralytics, 2023.
[18] Tsung-Yi Lin, Michael Maire, Serge Belongie, Lubomir Bourdev, Ross Girshick, James Hays, Pietro Perona, Deva Ramanan, C. Lawrence Zitnick, and Piotr Dollár. Microsoft COCO: Common objects in context. In Proceedings of the European Conference on Computer Vision (ECCV), pages 740–755, 2014.
[19] Dylan Lester, Pingping Zhu, Husnu Saner Narman, and Ammar Alzarrad. A YOLO-based semi-automated labeling approach to improve fault detection efficiency in railroad videos. In 2025 ASEE North Central Section (NCS) Annual Conference, Huntington, West Virginia, 2025. ASEE Conferences.
[20] Ming-Yueh Wang, Ying-Chieh Liu, and Jhih-Chuan Jheng. Development of an analytic model for the first three hits during table tennis match. Sports & Exercise Research, 14(2):229–238, 2012.
[21] Zheng Zhou and Hui Zhang. A visible analysis approach for table tennis tactical benefit. Journal of Sports Science and Medicine, 21(4):517–527, 2022.
[22] André Berchtold and Adrian E. Raftery. The mixture transition distribution (MTD) model for high-order Markov chains and non-gaussian time series. Technical Report 360, Department of Statistics, University of Washington, 1999.
[23] Ming-Hua Hsu, Cheng-Huan Yu, Ying-Lin Hsu, Ya-Lun Tsai, and Sheng-Kuang Wu. Analysis of receiving techniques and tactics of top mixed doubles table tennis players based on the 3S principle and machine learning: A case study of yun-ju lin/i-ching cheng. Sports & Exercise Research, 24(4):563–583, 2022.
[24] André Berchtold and Adrian E. Raftery. The mixture transition distribution model for high-order Markov chains and non-gaussian time series. Statistical Science, 17(3):328–356, 2002.
[25] André Berchtold, Ogier Maitre, and Kevin Emery. Optimization of the mixture transition distribution model using the march package for R. Symmetry, 12(12):2031, 2020.
[26] Lucio Caprioli, Amani Najlaoui, Francesca Campoli, Aatheethyaa Dhanasekaran, Saeid Edriss, Cristian Romagnoli, Andrea Zanela, Elvira Padua, Vincenzo Bonaiuto, and Giuseppe Annino. Tennis timing assessment by a machine learning-based acoustic detection system: A pilot study. Journal of Functional Morphology and Kinesiology, 10(1):47, 2025.