Scholarlink

Retrieving similar plays is fundamental to soccer analytics, but the lack of labeled data makes evaluation challenging. We study soccer play retrieval using ball trajectory data from the FIFA World Cup 2022 dataset. We examine Dynamic Time Warping (DTW) and Subsequence DTW, identifying limitations in computational scalability and degenerate alignment patterns. We propose an anchor-based representation that extracts sparse, tactically significant events and applies Subsequence DTW at the anchor level, reducing sequence length by 40%. We evaluate using data augmentation to generate perturbed versions of each play. The anchor-based method achieves 98% recall compared to 42% for full-trajectory DTW, demonstrating substantially improved robustness to sampling variability while maintaining 82.7% zonal precision in spatial context preservation.

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

Trajectory-Based Retrieval of Similar Soccer Plays Using Time-Warping and Anchor-Based Representations

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shehap Elhadary,
  • Amrou Abdelmawla,
  • Mahmoud Nasr,
  • Mohamed Hafez,
  • Tamer Basha

摘要

Retrieving similar plays is fundamental to soccer analytics, but the lack of labeled data makes evaluation challenging. We study soccer play retrieval using ball trajectory data from the FIFA World Cup 2022 dataset. We examine Dynamic Time Warping (DTW) and Subsequence DTW, identifying limitations in computational scalability and degenerate alignment patterns. We propose an anchor-based representation that extracts sparse, tactically significant events and applies Subsequence DTW at the anchor level, reducing sequence length by 40%. We evaluate using data augmentation to generate perturbed versions of each play. The anchor-based method achieves 98% recall compared to 42% for full-trajectory DTW, demonstrating substantially improved robustness to sampling variability while maintaining 82.7% zonal precision in spatial context preservation.