Space evaluation based on pitch control using drone video in Ultimate

Shunsuke Iwashita, Atom Scott, Rikuhei Umemoto, Ning Ding, Keisuke FujiiarXiv 2024 · DOI · arXiv

Abstract Quantitative analysis in Ultimate Frisbee has historically lagged behind other field sports due to its unique spatial dynamics. This study addresses this gap by proposing the Ultimate Scoring Opportunity (USO) index. By modifying soccer-based Potential Pitch Control Field (PPCF) models to reflect Ultimate’s specific rules—specifically the stationary thrower and stalling mechanics—the authors provide a framework for evaluating offensive space creation. Analysis of drone-captured 3-on-3 matches reveals that successful scoring is characterized by a significant rise in USO values and higher passing precision relative to optimal spatial zones.

Introduction

In sports like soccer or basketball, space is a fluid concept where every player is constantly moving. Ultimate presents a unique spatial paradox: the player holding the disc (the "thrower") must remain stationary, making the movement of the other six teammates the primary means of advancing the disc. Despite this tactical depth, there is a paucity of quantitative space evaluation in the sport. This research adapts computational sports analytics to Ultimate, providing a new framework for how players create and exploit "hot zones" on the field.

Methodology: Adapting Pitch Control

The researchers developed the Ultimate Scoring Opportunity (USO) index by adapting the Potential Pitch Control Field (PPCF), a concept from soccer that calculates the probability of a player controlling a specific point based on position and velocity. To align with Ultimate's rules, three critical modifications were made:

  1. Thrower Exclusion: The stationary thrower is excluded from space competition calculations.
  2. Marker Exclusion: Defenders within three meters of the thrower (performing a "stall") are excluded, as they are not covering open territory.
  3. Weighted Valuation: The model applies position weights ($w_{area}$) for end-zone proximity and distance weights ($w_{distance}$) to account for throw difficulty.
Comparison of the standard PPCF model (a) and the version adapted for Ultimate's stationary thrower rules (b).

Data Collection and Analysis

The study utilized drone footage of 3-on-3 Ultimate matches captured on a 54m x 20m court. Researchers manually annotated 262 video segments, using the Sportslabkit library to perform homography transformations, converting 2D video coordinates into real-world court coordinates. This allowed for a precise comparison between "Scoring" sets and "Turnover" sets.

Key Results

The USO metric successfully distinguished between high-quality offensive flow and stagnant play. In scoring sets, the average USO score rose from 0.536 to 0.597 in the 30 frames leading up to the final pass, whereas turnover sets remained stagnant around 0.518.

Furthermore, the data highlighted a significant "accuracy gap." In successful scoring sets, the distance between the pass destination and the optimal (highest USO) point was only 3.8 meters. In turnover sets, this distance more than doubled to 8.6 meters, suggesting that turnovers often occur when throwers attempt to hit sub-optimal or "closed" spaces.

Quantitative comparison of USO metrics and passing precision between scoring and turnover sets across the final 30 frames.

Conclusion and Future Work

The USO index provides a robust quantitative tool for evaluating spatial tactics in Ultimate. While the current study focused on 3-on-3 play, future work aims to scale the model to the standard 7-on-7 format. Additionally, the researchers identified the need for improved computer vision to automate player and disc tracking, which would allow for larger-scale analysis of defensive structures and offensive efficiency.

Summary updated

BibTeX
@misc{iwashita2024space,
  author = {Shunsuke Iwashita and Atom Scott and Rikuhei Umemoto and Ning Ding and Keisuke Fujii},
  title = {{Space evaluation based on pitch control using drone video in Ultimate}},
  year = {2024},
  eprint = {2409.14588},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2409.14588},
}
Figure