Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Contextual Sprint Classification in Soccer Based on Deep Learning

Hyunsung Kim, Gun-Hee Joe, Jinsung Yoon, Sang-Ki Ko

TL;DR

Experimental results show that the proposed deep learning framework for automatically classifying sprints in soccer into contextual categories has the potential for facilitating the integrated analysis of soccer sprints at scale.

Abstract

The analysis of high-intensity runs (or sprints) in soccer has long been a topic of interest for sports science researchers and practitioners. In particular, recent studies suggested contextualizing sprints based on their tactical purposes to better understand the physical-tactical requirements of modern match-play. However, they have a limitation in scalability, as human experts have to manually classify hundreds of sprints for every match. To address this challenge, this paper proposes a deep learning framework for automatically classifying sprints in soccer into contextual categories. The proposed model covers the permutation-invariant and sequential nature of multi-agent trajectories in soccer by deploying Set Transformers and a bidirectional GRU. We train the model with category labels made through the collaboration of human annotators and a rule-based classifier. Experimental results show that our model classifies sprints in the test dataset into 15 categories with the accuracy of 77.65%, implying the potential of the proposed framework for facilitating the integrated analysis of soccer sprints at scale.

Contextual Sprint Classification in Soccer Based on Deep Learning

TL;DR

Experimental results show that the proposed deep learning framework for automatically classifying sprints in soccer into contextual categories has the potential for facilitating the integrated analysis of soccer sprints at scale.

Abstract

The analysis of high-intensity runs (or sprints) in soccer has long been a topic of interest for sports science researchers and practitioners. In particular, recent studies suggested contextualizing sprints based on their tactical purposes to better understand the physical-tactical requirements of modern match-play. However, they have a limitation in scalability, as human experts have to manually classify hundreds of sprints for every match. To address this challenge, this paper proposes a deep learning framework for automatically classifying sprints in soccer into contextual categories. The proposed model covers the permutation-invariant and sequential nature of multi-agent trajectories in soccer by deploying Set Transformers and a bidirectional GRU. We train the model with category labels made through the collaboration of human annotators and a rule-based classifier. Experimental results show that our model classifies sprints in the test dataset into 15 categories with the accuracy of 77.65%, implying the potential of the proposed framework for facilitating the integrated analysis of soccer sprints at scale.
Paper Structure (12 sections, 3 equations, 6 figures, 2 tables)

This paper contains 12 sections, 3 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Detecting Sprints from Soccer Match Data
  4. Quantifying the Taxonomy and a Rule-Based Classifying Engine
  5. Constructing a Deep Learning Classifier
  6. Experiments
  7. Annotation of Ground Truth Labels
  8. Performance of the Deep Learning Classifier
  9. Use Cases
  10. Aggregation of Categorized Sprints
  11. Keyword-Guided Similar Play Retrieval
  12. Conclusions

Figures (6)

  • Figure 1: Speed plot of a player in a match. The dotted vertical lines indicate the borderlines between every deceleration and the following acceleration, among which selected cut-off points of run efforts are colored in red. The shaded area is a sprint whose peak speed exceeds 21kmh (i.e., the yellow horizontal line).
  • Figure 2: Instances of sprint categories. Note that the sprinter's team (colored in red) plays from left to right in every instance.
  • Figure 3: Distribution of 18 roles' mean locations per role period resulting from applying SoccerCPD to our dataset. The letters 'L', 'C', and 'R' in the front of labels stand for 'left', 'central', and 'right', respectively. Also, '(W)B', '(D/A)M', and 'F' on the latter part signify '(wing-)back', '(defensive/attacking) midfielder', and 'forward', respectively. For instance, 'LCB' means 'left center back'.
  • Figure 4: Confusion matrix for our deep learning classifier.
  • Figure 5: Role-by-role sprint counts of a match discretized by the true and predicted categories, respectively.
  • ...and 1 more figures

Theorems & Definitions (7)

  • Definition 1: Goal side
  • Definition 2: Target opponent
  • Definition 3: Passing line
  • Definition 4: Offside line
  • Definition 5: Defensive line
  • Definition 6: Defensive area
  • Definition 7: Return to defense