本論文提出了一種用於分析羽球比賽的低成本系統。該系統利用多個攝影機的視覺數據來追蹤羽球的 3D 軌跡和 2D 落地位置，以改善決策。該系統解決了環境變化、高速物件追蹤和多攝影機協調等關鍵挑戰，提高了自動化分析的準確性和可靠性。我們開發了一種基於關鍵點邊線交會點的新球場偵測方法，該方法包括兩個模型和一個映射演算法，改進了球場邊界的檢測。為了分析羽球的軌跡，我們對 TrackNetV2 模型進行了微調，應用三角測量進行 3D 重建，並進行後處理以提高準確性。此外，我們還設計了一種確定獲勝者和死球偵測的方法，以可靠地確定回合的獲勝者。透過將這些組件整合到一個綜合系統中，我們可以減少人為錯誤並提高效率。我們也收集自己的資料集來評估系統。實驗結果證明了系統的高效能，與手動檢測相比，大大減少了錯誤。透過簡化工作流程並節省體育產業的寶貴資源，這項研究將成為未來低成本電子裁判解決方案的墊腳石。;This thesis presents a low-cost system for analyzing badminton rallies. Utilizing visual data from multiple cameras, the system tracks the shuttlecock′s 3D trajectory and 2D landing positions to improve decision-making. Addressing key challenges such as environmental variability, high-speed object tracking, and multi-camera coordination, the system enhances the accuracy and reliability of automated analysis. We develop a new court detection method based on keypoint intersections, which includes two models and a mapping algorithm, improving the detection of court boundaries. To analyze the shuttlecock′s trajectory, we fine-tune the TrackNetV2 model, apply triangulation for 3D reconstruction, and perform post-processing for enhanced accuracy. Additionally, we devise a method for winner determination and dead ball detection to reliably determine the winner of a rally. By integrating these components into a comprehensive system, we reduce human errors and increase efficiency. We also collect our own datasets to evaluate the system. Experimental results demonstrate the system′s high performance, significantly reducing errors compared to manual detection. By streamlining workflows and conserving valuable resources within the sports industry, this research will be the stepping stone for future low-cost electronic refereeing solutions.
