Velocity Completion Task and Method for Event-based Player Positional Data in Soccer

TL;DR

The paper tackles the challenge that event-based soccer data lack continuous velocity information, hindering dynamic analyses. It defines a Velocity Completion Task and introduces a Graph Recurrent Neural Network (GRNN) that uses graph structure and temporal context to infer all players’ velocities from event-time positions. Empirical results show GRNN substantially outperforms a rule-based baseline in RMSE and yields space-evaluation metrics (PPCF and OBSO) that align more closely with full-tracking data, demonstrating practical value for advanced team-sport analytics. This approach enables velocity-aware analyses from publicly available event data, supporting more accurate assessments of player and team dynamics in real-world settings.

Abstract

In many real-world complex systems, the behavior can be observed as a collection of discrete events generated by multiple interacting agents. Analyzing the dynamics of these multi-agent systems, especially team sports, often relies on understanding the movement and interactions of individual agents. However, while providing valuable snapshots, event-based positional data typically lacks the continuous temporal information needed to directly calculate crucial properties such as velocity. This absence severely limits the depth of dynamic analysis, preventing a comprehensive understanding of individual agent behaviors and emergent team strategies. To address this challenge, we propose a new method to simultaneously complete the velocity of all agents using only the event-based positional data from team sports. Based on this completed velocity information, we investigate the applicability of existing team sports analysis and evaluation methods. Experiments using soccer event data demonstrate that neural network-based approaches outperformed rule-based methods regarding velocity completion error, considering the underlying temporal dependencies and graph structure of player-to-player or player-to-ball interaction. Moreover, the space evaluation results obtained using the completed velocity are closer to those derived from complete tracking data, highlighting our method's potential for enhanced team sports system analysis.

Figures (7)

• Figure 1: Flow chart of the task framework proposed in this study. We first extracted the coordinates of the ball and players from the tracking data and event data. We then input them into the model, and output the complemented player velocities.
• Figure 2: The GRNN model proposed in this study. If we also used the event action type ($a^{t}$) and ball coordinates at the end of the event ($x^{t}_{\mathrm{end}}$) as additional features, we selected $\mathrm{GRNN_{add}}$; otherwise, we utilized GRNN.
• Figure 3: Results of comparing the distribution of predicted player velocities by GNN, RNN, $\mathrm{GRNN}$, and $\mathrm{GRNN_{dec\_add}}$ with the distribution of true velocities.
• Figure 4: Error distribution over the pitch for PPCF and OBSO calculated using velocities complemented by the rule-based baseline or GRNN compared to ground truth.
• Figure 5: Example heatmaps of PPCF computed using ground truth, rule-based, or GRNN-complemented velocities.