Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hand Held Multi-Object Tracking Dataset in American Football

Rintaro Otsubo, Kanta Sawafuji, Hideo Saito

TL;DR

This work tackles the lack of a standardized MOT benchmark for American football by introducing the first two-view, hand-held camera dataset with roughly 200k bounding boxes across over 10k frames. It evaluates a range of detection and tracking approaches, showing that fine-tuning detectors and ReID models leads to substantial improvements in both detection and tracking accuracy, with BoT-SORT plus a fine-tuned detector delivering the best overall MOT performance. The study highlights the pivotal role of accurate detection in occlusion-heavy football scenes and demonstrates that OSNet can outperform transformer-based ReID in this domain. Overall, the dataset and experimental findings provide a valuable resource for device-free, vision-based analysis of football games and a benchmark for cross-domain MOT research.

Abstract

Multi-Object Tracking (MOT) plays a critical role in analyzing player behavior from videos, enabling performance evaluation. Current MOT methods are often evaluated using publicly available datasets. However, most of these focus on everyday scenarios such as pedestrian tracking or are tailored to specific sports, including soccer and basketball. Despite the inherent challenges of tracking players in American football, such as frequent occlusion and physical contact, no standardized dataset has been publicly available, making fair comparisons between methods difficult. To address this gap, we constructed the first dedicated detection and tracking dataset for the American football players and conducted a comparative evaluation of various detection and tracking methods. Our results demonstrate that accurate detection and tracking can be achieved even in crowded scenarios. Fine-tuning detection models improved performance over pre-trained models. Furthermore, when these fine-tuned detectors and re-identification models were integrated into tracking systems, we observed notable improvements in tracking accuracy compared to existing approaches. This work thus enables robust detection and tracking of American football players in challenging, high-density scenarios previously underserved by conventional methods.

Hand Held Multi-Object Tracking Dataset in American Football

TL;DR

This work tackles the lack of a standardized MOT benchmark for American football by introducing the first two-view, hand-held camera dataset with roughly 200k bounding boxes across over 10k frames. It evaluates a range of detection and tracking approaches, showing that fine-tuning detectors and ReID models leads to substantial improvements in both detection and tracking accuracy, with BoT-SORT plus a fine-tuned detector delivering the best overall MOT performance. The study highlights the pivotal role of accurate detection in occlusion-heavy football scenes and demonstrates that OSNet can outperform transformer-based ReID in this domain. Overall, the dataset and experimental findings provide a valuable resource for device-free, vision-based analysis of football games and a benchmark for cross-domain MOT research.

Abstract

Multi-Object Tracking (MOT) plays a critical role in analyzing player behavior from videos, enabling performance evaluation. Current MOT methods are often evaluated using publicly available datasets. However, most of these focus on everyday scenarios such as pedestrian tracking or are tailored to specific sports, including soccer and basketball. Despite the inherent challenges of tracking players in American football, such as frequent occlusion and physical contact, no standardized dataset has been publicly available, making fair comparisons between methods difficult. To address this gap, we constructed the first dedicated detection and tracking dataset for the American football players and conducted a comparative evaluation of various detection and tracking methods. Our results demonstrate that accurate detection and tracking can be achieved even in crowded scenarios. Fine-tuning detection models improved performance over pre-trained models. Furthermore, when these fine-tuned detectors and re-identification models were integrated into tracking systems, we observed notable improvements in tracking accuracy compared to existing approaches. This work thus enables robust detection and tracking of American football players in challenging, high-density scenarios previously underserved by conventional methods.

Paper Structure

This paper contains 16 sections, 5 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Detecting and Tracking Players
  4. Multi-Object Tracking Datasets
  5. Person Re-Identification
  6. Dataset
  7. Data Collection
  8. Annotation Process
  9. Dataset Statistics and Analysis
  10. Experiments
  11. Experimental Setup
  12. Implementation Details
  13. Detection
  14. Tracking
  15. Ablation Study
  16. ...and 1 more sections

Figures (5)

  • Figure 1: Examples of play categories by type (rows: pass, run, punt) and phase (columns: huddle to near-end). A ball or key players holding the ball are highlighted as an orange bounding box for clarity.
  • Figure 2: Total number of annotations per opponent university. Orange bars represent wide, and blue bars represent tight.
  • Figure 3: Average number of annotations per opponent university. Orange bars represent wide, and blue bars represent tight.
  • Figure 4: Tracklet length histogram per opponent university. Orange bars represent wide, and blue bars represent tight.
  • Figure 5: Qualitative Results on Detect players and referees