Tracking of Ball and Players in Beach Volleyball Videos

Gabriel J. García, Patricia Herrera López, Daniel Link, Bjoern M. EskofierPLoS 2014 · DOI

Abstract

This paper presents methods for the determination of players' positions and contact time points by tracking the players and the ball in beach volleyball videos. Two player tracking methods are compared: a classical particle filter and a rigid grid integral histogram tracker. Due to mutual occlusion of the players and the camera perspective, results are best for the front players, with 74.6% and 82.6% of correctly tracked frames for the particle method and the integral histogram method, respectively. Results suggest an improved robustness against player confusion between different particle sets when tracking with a rigid grid approach. Faster processing and less player confusion make this method superior to the classical particle filter. Two different ball tracking methods are used that detect ball candidates from movement difference images using a background subtraction algorithm. Ball trajectories are estimated and interpolated from parabolic flight equations. The tracking accuracy of the ball is 54.2% for the trajectory growth method and 42.1% for the Hough line detection method. Tracking results of over 90% from the literature could not be confirmed. Ball contact frames were estimated from parabolic trajectory intersection, resulting in 48.9% of correctly estimated ball contact points.

Introduction

Automated sports analysis is a rapidly evolving field, yet beach volleyball presents unique "in-the-wild" challenges that complicate standard computer vision techniques. Unlike the controlled environments of indoor courts, beach volleyball involves visual noise from sand textures mimicking skin tones, dynamic lighting shifts, and complex occlusions as players cross paths or are obscured by the net. Furthermore, the high velocity of the ball often results in motion blur in standard 25-30 fps video. This study introduces "BeachTracker," a system designed to provide trainers with automated statistics on player positioning and ball contact points by overcoming these environmental hurdles.

Methodology

The researchers developed a multi-stage pipeline focusing on two primary tasks: player tracking and ball tracking.

Player Tracking

The system employs a Homographic Transformation to map 2D video pixels into 3D world coordinates. Tracking is driven by three weighted cues:

  1. Foreground Mask: An adaptive Gaussian Mixture Model (GMM) isolates moving objects.
  2. Movement Weighting: Identifying pixel changes between consecutive frames.
  3. Color Weighting: Utilizing HSV histograms and the Bhattacharyya distance to differentiate players.

Two algorithmic approaches were compared: a Classical Particle Filter, which uses a stochastic cloud of particles that resamples based on probability, and a Rigid Grid Integral Histogram tracker. The latter maintains a fixed grid of particles around the player and utilizes integral images for rapid histogram calculation.

Ball Tracking and Contact Detection

Ball candidates are identified via movement difference images. The system enforces physical constraints, assuming a linear horizontal path and a parabolic vertical path. The study compared a "trajectory growth" method against a newly proposed method based on Hough line detection. Ball contact points were estimated by identifying intersections in the parabolic trajectories.

Datasets

Evaluation was conducted on manually annotated video data (640x480 resolution):

  • Player Tracking: 12 videos (avg. 259 frames) of male and female matches.
  • Ball Tracking: 7 videos (535 frames).
  • Contact Points: 31 videos (190 ball contacts).

Results and Discussion

The Rigid Grid Integral Histogram method proved superior for player tracking, achieving 82.6% accuracy for front players compared to 74.6% for the particle filter. This approach was also approximately twice as fast, making it more viable for near-real-time applications. The rigid grid prevents "particle split," a common failure in traditional filters where the particle cloud expands and accidentally tracks the wrong player.

Ball tracking proved significantly more difficult. The trajectory growth method reached 54.2% accuracy, while the Hough-based method achieved only 42.1%. Notably, the authors were unable to replicate the 90%+ accuracy rates reported in existing literature, suggesting that previous studies may have relied on "favorable" video segments rather than the noisy, real-world conditions of competitive matches. Ball contact point estimation reached an accuracy of 48.9% within a 10-frame window.

Conclusion

The study demonstrates that a rigid grid approach offers improved robustness and efficiency for player tracking in beach volleyball. However, ball tracking remains a significant challenge due to camera perspective and motion blur. Future improvements may involve higher-definition video, side-court camera placement to reduce occlusions, and the development of confidence measures to flag frames for manual human review.

Summary updated

Abstract

This paper presents methods for the determination of players' positions and contact time points by tracking the players and the ball in beach volleyball videos. Two player tracking methods are compared, a classical particle filter and a rigid grid integral histogram tracker. Due to mutual occlusion of the players and the camera perspective, results are best for the front players, with 74,6% and 82,6% of correctly tracked frames for the particle method and the integral histogram method, respectively. Results suggest an improved robustness against player confusion between different particle sets when tracking with a rigid grid approach. Faster processing and less player confusions make this method superior to the classical particle filter. Two different ball tracking methods are used that detect ball candidates from movement difference images using a background subtraction algorithm. Ball trajectories are estimated and interpolated from parabolic flight equations. The tracking accuracy of the ball is 54,2% for the trajectory growth method and 42,1% for the Hough line detection method. Tracking results of over 90% from the literature could not be confirmed. Ball contact frames were estimated from parabolic trajectory intersection, resulting in 48,9% of correctly estimated ball contact points.
BibTeX
@inproceedings{garcia2014tracking,
  author = {Gabriel J. García and Patricia Herrera López and Daniel Link and Bjoern M. Eskofier},
  title = {{Tracking of Ball and Players in Beach Volleyball Videos}},
  booktitle = {PLoS},
  year = {2014},
  doi = {10.1371/journal.pone.0111730},
}
Figure