運動科技在市場上已逐漸受到關注，尤其棒球在台灣普遍認為是國球，不僅是場上參與的球員和教練組，場下也有龐大的觀眾。在棒球比賽中，投手通常擔綱比賽最重要的位置，有著好的投球內容能帶領球隊勝利，不過裁判對於好球帶的判決卻十分主觀。現在在各大層級聯賽逐漸採用電子好球帶作為輔助，以提供觀眾好壞球的資訊，然則這方面的設備相對昂貴和不便裝置，因此本論文設計一套演算法，使用傳統影像處理結合機器學習的結果，做出僅用一顆攝影機的好壞球辨識系統。
在本論文中，電子好球帶通常於固定尺寸，計算三維空間和二維空間的投影矩陣，將三維空間的本壘板投影至利用影像辨識出的本壘板，進而得到好球帶位置。設備上使用一台一般手機錄製投手投球影片，再匯入電腦剪輯出每一次投球。針對每一次投球以YOLOv4-tiny作為模型進行人物辨識，再以傳統影像處理方式追蹤棒球並畫出投球軌跡，再根據投球軌跡進行預測棒球位置，最終使用速率公式推測進壘的時間點，進而判斷好壞球與否。本實驗於實際的棒球場配合校隊判斷來收集資料，以此評估系統的有效性，實驗結果顯示此系統在辨別好壞球上達到88%的準確度。最後相信本論文的研究對於運動科技的普及和發展能做出幫助。

Sports technology has gradually garnered attention in the market, especially in Taiwan where baseball is widely regarded as the national sport. This sport involves not only the players and coaching staff on the field but also a substantial audience off the field. In baseball games, pitchers typically play a crucial role, and their pitching performance significantly influences the team's success. However, the umpire's judgment of the strike zone is often highly subjective. This thesis proposes an algorithm that combines traditional image processing with machine learning to create a ball and strike recognition system using only a single camera.

In this thesis, the electronic strike zone is typically of a fixed size. The projection matrix for both three-dimensional and two-dimensional spaces is calculated, projecting the three-dimensional space of the home plate onto the home plate recognized using image recognition, thus determining the position of the strike zone. The equipment involves using a standard smartphone to record pitching videos, which are then imported into a computer and edited for each pitching instance. For each pitch, YOLOv4-tiny is employed as the model for person detection, followed by traditional image processing to track the baseball and plot its trajectory. Based on the predicted baseball trajectory, the system estimates the baseball's position and uses the velocity formula to infer the time of reaching home plate, thereby determining whether it is a strike or a ball. The experiment was conducted on an actual baseball field in collaboration with a collegiate baseball team to collect data, and the results showed a 88% accuracy rate in distinguishing between balls and strikes.

中文摘要 .................................................................................................................................... i
Abstract .................................................................................................................................... ii
Contents ................................................................................................................................... iii
Chapter 1 Introduction ............................................................................................................... 1
1.1 Motivation and Objective ......................................................................................... 1
1.2 Thesis Organization ................................................................................................. 2
Chapter 2 Background ............................................................................................................... 3
2.1 YOLOv4................................................................................................................... 3
2.1.1 YOLOv4-tiny ....................................................................................................... 6
2.2 Traditional Image Processing ................................................................................... 6
2.2.1 Image Contrast ..................................................................................................... 6
2.2.2 Binarization.......................................................................................................... 7
2.2.3 Image Contour Detection ..................................................................................... 7
2.2.4 Canny Edge Detection ......................................................................................... 8
2.3 Camera Projection .................................................................................................... 9
Chapter 3 Proposed Methodology ........................................................................................... 10 3.1 System Introduction ............................................................................................... 10 3.2 Detecting the Home Plate....................................................................................... 11 3.3 Establishing the Strike Zone .................................................................................. 12 3.4 Baseball Tracking .................................................................................................. 15 3.4.1 Pitching Moment ............................................................................................... 15 3.4.2 The Trajectory of The Baseball ......................................................................... 16
Chapter 4 Experimental Design and Results ........................................................................... 18 4.1 Experimental Environment Setup .......................................................................... 18
iv
4.2 Experimental Procedure ......................................................................................... 19 4.2.1 Calculation The Time of Ball Passing The Strike Zone .................................... 19 4.2.2 Labeling ............................................................................................................. 20 4.3 Experiment Result .................................................................................................. 21 4.3.1 Judgment Results ............................................................................................... 21 4.3.2 Execution Time .................................................................................................. 24
Chapter 5 Conclusions and Future Work ................................................................................ 26
5.1 Conclusions ............................................................................................................ 26
5.2 Future Work ........................................................................................................... 26
Bibliography ............................................................................................................................ 28

[1] Guéziec, André. "Tracking pitches for broadcast television." Computer 35.3 (2002): 38-43.
[2] Mike Fast, “What the heck is PITCH f/x?,” The Hardball Times Baseball Annual (2010): 153-158.
[3] J. Martin, “Evaluation of doppler radar ball tracking and its experimental uses,” Ph.D. dissertation, Dept. Mech. Eng., Washington State Univ., WA, USA, 2012.
[4] H. Lee, J. Kim, J. Kim, W. Y. Kim, "A Method of Measuring Baseball Position at the Strike Zone." 2020 International Conference on Electronics, Information, and Communication (ICEIC). IEEE, 2020.
[5] R. Cavallaro, M. Hybinette, M. White, T. Balch “Augmenting Live Broadcast Sports with 3D Tracking Information,” IEEE MultiMedia 18, no. 4 (2011): 38-47.
[6] C.-G Cho, “Screen baseball game system and the method using infraredcamera bar,” K.R. Patent KR101548511B1, Sep. 1, 2015.
[7] Y,-P. Chen ”A Baseball Strike and Ball Recognition System Based on Deep Learning” master's thesis, Dept. Computer Science & Information Engineering, National Central University, TW, 2022。
[8] Y. Yamaguchi, & M. Motoki. "Real-time analysis of baseball pitching using image processing on smartphone." Procedia Computer Science 96 (2016): 1059-1066.
[9] A. Bochkovskiy, C.-Y. Wang, & H.-Y. M. Liao, “YOLOv4: Optimal Speed and Accuracy of Object Detection,” arXiv, Tech. Rep. arXiv:2004.10934, Apr. 2020.
[10] C.-Y. Wang, H.-Y. Mark Liao, Y.-H. Wu, P.-Y. Chen, & J.-W. Hsieh, and I.-H. Yeh, “CSPNet:A New Backbone that can Enhance Learning Capability of CNN,” in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Seattle,WA, USA: IEEE, Jun. (2020): 1571–1580.
[11] S. Liu, L. Qi, H. Qin, J. Shi, a& J. Jia, “Path Aggregation Network for Instance Segmentation,” arXiv, Tech. Rep. arXiv:1803.01534, Sep. 2018.
[12] K. He, X. Zhang, S. Ren, an& J. Sun, “Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition,” arXiv, Tech. Rep. arXiv:1406.4729, Apr. 2015.
[13] Z. Jiang, L. Zhao, S. Li, & Y. Jia. "Real-time object detection method based on improved YOLOv4-tiny." arXiv preprint arXiv:2011.04244 ,2020.
[14] C. -Y. Wang, Alexey Bochkovskiy, & H. -Y. Liao. "Scaled-yolov4: Scaling cross stage partial network." Proceedings of the IEEE/cvf conference on computer vision and pattern recognition. 2021.
29
[15] W. Boyle, “Prospectus Feature: The Universal Strike Zone” Baseball Prospectus ,2018.