Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards long-term player tracking with graph hierarchies and domain-specific features

Maria Koshkina, James H. Elder

TL;DR

SportsSUSHI introduces a hierarchical graph-based tracker for long-term player tracking in team sports, leveraging domain-specific cues—jersey numbers, team IDs, and field coordinates—to maintain identities across occlusions and camera motion. Built on the SUSHI framework, it employs level-specific feature encoders within a multi-level graph to connect short tracklets into long trajectories without game-specific retraining. Evaluations on SoccerNet and a newly proposed hockey dataset show improvements in association accuracy and overall tracking metrics, underscoring the value of incorporating domain knowledge into offline MOT for sports. The work provides a publicly available dataset and code, facilitating further research in long-term sports tracking and cross-sport applicability.

Abstract

In team sports analytics, long-term player tracking remains a challenging task due to player appearance similarity, occlusion, and dynamic motion patterns. Accurately re-identifying players and reconnecting tracklets after extended absences from the field of view or prolonged occlusions is crucial for robust analysis. We introduce SportsSUSHI, a hierarchical graph-based approach that leverages domain-specific features, including jersey numbers, team IDs, and field coordinates, to enhance tracking accuracy. SportsSUSHI achieves high performance on the SoccerNet dataset and a newly proposed hockey tracking dataset. Our hockey dataset, recorded using a stationary camera capturing the entire playing surface, contains long sequences and annotations for team IDs and jersey numbers, making it well-suited for evaluating long-term tracking capabilities. The inclusion of domain-specific features in our approach significantly improves association accuracy, as demonstrated in our experiments. The dataset and code are available at https://github.com/mkoshkina/sports-SUSHI.

Towards long-term player tracking with graph hierarchies and domain-specific features

TL;DR

SportsSUSHI introduces a hierarchical graph-based tracker for long-term player tracking in team sports, leveraging domain-specific cues—jersey numbers, team IDs, and field coordinates—to maintain identities across occlusions and camera motion. Built on the SUSHI framework, it employs level-specific feature encoders within a multi-level graph to connect short tracklets into long trajectories without game-specific retraining. Evaluations on SoccerNet and a newly proposed hockey dataset show improvements in association accuracy and overall tracking metrics, underscoring the value of incorporating domain knowledge into offline MOT for sports. The work provides a publicly available dataset and code, facilitating further research in long-term sports tracking and cross-sport applicability.

Abstract

In team sports analytics, long-term player tracking remains a challenging task due to player appearance similarity, occlusion, and dynamic motion patterns. Accurately re-identifying players and reconnecting tracklets after extended absences from the field of view or prolonged occlusions is crucial for robust analysis. We introduce SportsSUSHI, a hierarchical graph-based approach that leverages domain-specific features, including jersey numbers, team IDs, and field coordinates, to enhance tracking accuracy. SportsSUSHI achieves high performance on the SoccerNet dataset and a newly proposed hockey tracking dataset. Our hockey dataset, recorded using a stationary camera capturing the entire playing surface, contains long sequences and annotations for team IDs and jersey numbers, making it well-suited for evaluating long-term tracking capabilities. The inclusion of domain-specific features in our approach significantly improves association accuracy, as demonstrated in our experiments. The dataset and code are available at https://github.com/mkoshkina/sports-SUSHI.

Paper Structure

This paper contains 25 sections, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. MOT
  4. Player MOT
  5. Jersey Number Recognition
  6. Team Identification
  7. Field Registration
  8. Method
  9. SportsSUSHI
  10. Feature Extraction
  11. Jersey Number Prediction
  12. Team ID Prediction
  13. Field Coordinates
  14. Person Re-ID
  15. Edge Features
  16. ...and 10 more sections

Figures (4)

  • Figure 1: Team uniform and protective gear make players on the same team hard to distinguish. Jersey number is a key feature for player re-identification and long-term player tracking.
  • Figure 2: SportsSUSHI: we propose a player tracking system based on hierarchical tracker SUSHICetintas_2023_CVPR. Our feature extraction module, extracts features crucial for player tracking: jersey number, field coordinates, and team ID, in addition to classic re-ID features. The tracker then builds a hierarchy of graphs where each next level spans longer temporal distances. Initial graphs contain detections as nodes. Similarity measures between node features serve as the edge feature. Each following layer contains tracklets formed by solving the graph in the previous step as nodes.
  • Figure 3: Sample frames from our hockey dataset.
  • Figure 4: Hockey dataset sequence length.