棒球被認為是台灣的國球，不管是場上的球員或是觀賽的球迷，有越來越多人開始投入這項運動。在棒球比賽中，投手扮演著非常重要的角色，一位好的投手可以主宰整場比賽的氣勢及節奏，然而裁判對於好球帶的判定是非常主觀的，往往成為影響比賽勝負的關鍵。有些職棒比賽採用電子好球帶輔助系統來提供投手好壞球資訊給觀眾球迷判定，但是在業餘比賽中會受限於場地限制或是設備較昂貴等因素而無法使用，因此本論文結合深度學習及影像處理技術做出類似但採用低成本的方式實做出一個好壞球辨識系統。
本論文之系統使用兩台低成本設備錄製投手投球影片，以YOLOv4模型進行棒球偵測，並且利用卡爾曼濾波器進行棒球軌跡預測及補影格。由於使用兩台攝影設備錄製影片，因此需要對影片做同步，本系統使用短時距傅立葉轉換擷取音訊特徵並且使用此特徵對齊影片時間。棒球比賽中的好球帶會依據本壘板位置及打者的身高做變化，本研究利用影像處理方式做本壘板偵測，並且使用OpenPose模型對打者進行骨架分析取得好球帶位置。
本研究利用實際至棒球場投球的資料集來評估本論文之系統有效性。根據實驗結果顯示，本系統在辨識好壞球時，準確率可以達到96%。

Baseball is considered the national sport of Taiwan, and there are more and more people committed to this sport, whether it is the players on the field or the fans watching the game. In a baseball game, the pitcher plays a major role who can dominate the momentum and rhythm of the game, but the decision of the referee on the strike zone is very subjective and often becomes the key to winning or losing the game. Some professional baseball games use electronic strike zone assistance systems to provide pitchers with strike and ball information to the fans, but this is impossible in amateur games due to venue constraints or expensive equipment. Therefore, this paper combines deep learning and image processing techniques to make a similar but low-cost way to implement a strike and ball recognition system.
The system in this paper uses two low-cost devices to record pitcher's pitching video, YOLOv4 model for baseball detection, and Kalman filter for baseball trajectory prediction and complementary frames. Since two cameras are used to record the video, it is necessary to synchronize the video. The system uses Short-time Fourier Transform to capture the audio features and uses this feature to align the video timing. The strike zone in a baseball game will change according to the position of the home plate and the height of the batter. This study uses image processing to detect the home plate and uses the OpenPose model to analyze the batter's skeleton to obtain the location of the strike zone.
This study uses the dataset which is actually pitching at the baseball field to evaluate the effectiveness of the system in this paper. According to the experimental results,
the accuracy of the system in identifying strike and ball can reach 96%.

[1] “The Impact of the Hawk-Eye System in Tennis.” (Jan. 2012), [Online]. Available: https://trainingwithjames.wordpress.com/research-papers/the- impact-of-the-hawk-eye-system-in-tennis/.
[2] A. Bochkovskiy, C.-Y. Wang, and H.-Y. M. Liao, “YOLOv4: Optimal Speed and Accuracy of Object Detection,” arXiv, Tech. Rep. arXiv:2004.10934, Apr. 2020.
[3] J. Redmon and A. Farhadi, “YOLOv3: An Incremental Improvement,” arXiv, Tech. Rep. arXiv:1804.02767, Apr. 2018.
[4] 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, pp. 1571–1580.
[5] T.-Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature Pyramid Networks for Object Detection,” arXiv, Tech. Rep. arXiv:1612.03144, Apr. 2017.
[6] S. Liu, L. Qi, H. Qin, J. Shi, and J. Jia, “Path Aggregation Network for Instance Segmentation,” arXiv, Tech. Rep. arXiv:1803.01534, Sep. 2018.
[7] K. He, X. Zhang, S. Ren, and J. Sun, “Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition,” arXiv, Tech. Rep. arXiv:1406.4729, Apr. 2015.
[8] S. Yun, D. Han, S. J. Oh, S. Chun, J. Choe, and Y. Yoo, “CutMix: Regularization Strategy
to Train Strong Classifiers with Localizable Features,” arXiv, Tech. Rep. arXiv:1905.04899, Aug. 2019.
[9] D. Misra, “Mish: A Self Regularized Non-Monotonic Activation Function,” arXiv, Tech. Rep. arXiv:1908.08681, Aug. 2020.
[10] T. Basar, “A new approach to linear filtering and prediction problems,” in Control The-
ory: Twenty-Five Seminal Papers (1932-1981), IEEE, 2001, pp. 167–179.
[11] S. Suzuki and K. Abe, “Topological structural analysis of digitized binary images by border following,” Comput. Vis. Graph. Image Process., vol. 30, pp. 32–46, 1985.
[12] D. Gabor, “Theory of communication. Part 1: The analysis of information,” Journal of the Institution of Electrical Engineers - Part III: Radio and Communication Engineering, vol. 93, no. 26, p. 429, 1946.
[13] Z. Cao, T. Simon, S.-E. Wei, and Y. Sheikh, “Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields,” arXiv, Tech. Rep. arXiv:1611.08050, Apr. 2017.
[14] Z. Cao, G. Hidalgo, T. Simon, S.-E. Wei, and Y. Sheikh, “OpenPose: Realtime MultiPerson 2D Pose Estimation using Part Affinity Fields,” arXiv, Tech. Rep. arXiv:1812.08008,
May 2019.
[15] K.-C. Huang, T.-K. Chiu, C.-T. Hsieh, and M.-S. Pan, “3D 深度感測技術應用在棒壘球好球帶輔助系統之開發,” zh, 淡江體育學刊, no. 20, pp. 1–14, Nov. 2017.
[16] “棒 球 規 則： 第 二 章-用 語 之 定 義 - 台 灣 棒 球 維 基 館,” [Online]. Available: http://twbsball.dils.tku.edu.tw/wiki/index.php/%E6%A3%92%E7%90%83%E8%A6%8F%E5%89%87_%EF%BC%9A%E7%AC%AC%E4%BA%8C%E7%AB%A0%E7%94%A8%E8%AA%9E%E4%B9%8B%E5%AE%9A%E7%BE%A9.
[17] C. Gaines. “What An MLB Strike Zone Really Looks Like And Why Players Are Always So Mad About It,” [Online]. Available: https://www.businessinsider.com/mlb-strike-zone-2014-9.
[18] T.-Y. Lin, M. Maire, S. Belongie, et al., “Microsoft COCO: Common Objects in Context,” arXiv, Tech. Rep., Feb. 2015.
[19] C. Corcoran. “New rule continues to be a big problem for MLB, umpires,” [Online]. Available: https://www.si.com/mlb/2014/08/14/home- plate- collision- rule- umpireswrong.
[20] C. Muller. ““Keep Yourself at 17 Inches＂.” (Mar. 2016), [Online]. Available: https://mygoodtimestories.com/2016/03/30/keep-yourself-at-17-inches/.
[21] “Why is home plate shaped the way it is?” [Online]. Available: https://www.theoleballgame.com/why-is-home-plate-shaped-the-way-it-is.html.