Scholarlink

This study presents a 3D baseball swing reconstruction system that estimates 3D swing trajectories based on synchronized dual-view videos. It is a pure vision-based method without the requirement of other sensors. 2D keypoints of the baseball bat are detected, and 3D swing trajectories are constructed based on associated keypoints from two views. Based on 3D swing trajectories, we calculate several swing metrics like attack angle and bat speed, and compare them with commercially available sensors to show the effectiveness of the proposed method. Furthermore, we introduce the BaseballSwing3D dataset that contains synchronized dual-view swing videos associated with 3D coordinates of the bat’s head point and the tail point. This dataset serves as a quantitative benchmark for validating the effectiveness of a 3D swing reconstruction method.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Vision-Based 3D Baseball Swing Trajectory Reconstruction and Swing Performance Analysis

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yi-Hsuan Lu,
  • Wei-Ta Chu

摘要

This study presents a 3D baseball swing reconstruction system that estimates 3D swing trajectories based on synchronized dual-view videos. It is a pure vision-based method without the requirement of other sensors. 2D keypoints of the baseball bat are detected, and 3D swing trajectories are constructed based on associated keypoints from two views. Based on 3D swing trajectories, we calculate several swing metrics like attack angle and bat speed, and compare them with commercially available sensors to show the effectiveness of the proposed method. Furthermore, we introduce the BaseballSwing3D dataset that contains synchronized dual-view swing videos associated with 3D coordinates of the bat’s head point and the tail point. This dataset serves as a quantitative benchmark for validating the effectiveness of a 3D swing reconstruction method.