We utilize characteristics of synchronized multi-view videos. Based on a pose detection model, we detect each player＇s position and key points. We then develop a multi-view multi-dimensional association algorithm to find correspondence of detections across different views. Then calculate 3D positions and perform 3D tracking, generating the 3D trajectories. Our system can be applied to tactical analysis, and player performance assessment, providing scientific data to players and coaches.

In academia, works on synchronized multi-view video data are still unexplored. Most existing multi-view datasets focus on improving 2D tracking and cross-scene tracking, rather than on reconstructing 3D trajectories using multi-view approaches. In practical applications, 3D trajectory tracking is highly valuable but less explored. This highlights the unique technical features of our research in multi-view, multi-player 3D tracking.

我們的技術可以用於排球、籃球、足球這類有球員動態的體育項目。只需架設攝影機在能拍攝到場地的位置，不需在球員身上穿戴設備或標記，即可追蹤所有球員的3D軌跡。隨著深度學習技術進步，未來有望用於例如辨識進攻戰術或分析敵隊行為模式，幫助球隊調整策略。我們的技術為運動競賽數據化提供基礎，具有廣泛應用潛力。