Spatial pixels selection and inter-frame combined likelihood based observation for 60 fps 3d tracking of twelve volleyball players on gpu

Yiming Zhao, Xina Cheng, Takeshi Ikenaga

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Citation (Scopus)

Abstract

3D players tracking plays an important role in sports analysis. Tracking of players contributes to high level game analysis such as tactic analysis and commercial applications such as TV contents. Many services like sports live and broadcasting have strict limitation on processing time, thus real-time implementation for 3D players tracking is necessary. This paper proposes a particle filter based 60 fps multi-view volleyball players tracking system on GPU platform. There are three proposals: body region constraint prediction, spatial pixels selection and inter-frame combined likelihood. The body region constraint prediction uses player’s body region as limitation in prediction to increase tracking accuracy. The spatial pixels selection method selects pixels for likelihood calculating to reduce calculation amount in spatial space. The inter-frame observation method does particle filter algorithm with two frames each time to reduce calculation amount in temporal space. Our experiments are based on videos of the Final and Semi-Final Game of 2014 Japan Inter High School Games of Men’s Volleyball in Tokyo Metropolitan Gymnasium. On the GPU device GeForce GTX 1080Ti, our tracking system achieves real-time on 60 fps videos and keeps the tracking accuracy higher than 97%.

Original languageEnglish
Title of host publicationAdvances in Multimedia Information Processing – PCM 2018 - 19th Pacific-Rim Conference on Multimedia, 2018, Proceedings
EditorsWen-Huang Cheng, Toshihiko Yamasaki, Chong-Wah Ngo, Richang Hong, Meng Wang
PublisherSpringer-Verlag
Pages716-726
Number of pages11
ISBN (Print)9783030007669
DOIs
Publication statusPublished - 2018 Jan 1
Event19th Pacific-Rim Conference on Multimedia, PCM 2018 - Hefei, China
Duration: 2018 Sep 212018 Sep 22

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11165 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other19th Pacific-Rim Conference on Multimedia, PCM 2018
CountryChina
CityHefei
Period18/9/2118/9/22

Fingerprint

Likelihood
Pixel
Pixels
Sports
Particle Filter
Tracking System
Game
Real time systems
Broadcasting
Spatial Prediction
Real-time
Prediction
Japan
Processing
High Accuracy
Observation
Experiments
Necessary
Graphics processing unit
Experiment

Keywords

  • 3D volleyball players tracking
  • GPU acceleration
  • Particle filter
  • Sports analysis

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Zhao, Y., Cheng, X., & Ikenaga, T. (2018). Spatial pixels selection and inter-frame combined likelihood based observation for 60 fps 3d tracking of twelve volleyball players on gpu. In W-H. Cheng, T. Yamasaki, C-W. Ngo, R. Hong, & M. Wang (Eds.), Advances in Multimedia Information Processing – PCM 2018 - 19th Pacific-Rim Conference on Multimedia, 2018, Proceedings (pp. 716-726). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11165 LNCS). Springer-Verlag. https://doi.org/10.1007/978-3-030-00767-6_66

Spatial pixels selection and inter-frame combined likelihood based observation for 60 fps 3d tracking of twelve volleyball players on gpu. / Zhao, Yiming; Cheng, Xina; Ikenaga, Takeshi.

Advances in Multimedia Information Processing – PCM 2018 - 19th Pacific-Rim Conference on Multimedia, 2018, Proceedings. ed. / Wen-Huang Cheng; Toshihiko Yamasaki; Chong-Wah Ngo; Richang Hong; Meng Wang. Springer-Verlag, 2018. p. 716-726 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11165 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Zhao, Y, Cheng, X & Ikenaga, T 2018, Spatial pixels selection and inter-frame combined likelihood based observation for 60 fps 3d tracking of twelve volleyball players on gpu. in W-H Cheng, T Yamasaki, C-W Ngo, R Hong & M Wang (eds), Advances in Multimedia Information Processing – PCM 2018 - 19th Pacific-Rim Conference on Multimedia, 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11165 LNCS, Springer-Verlag, pp. 716-726, 19th Pacific-Rim Conference on Multimedia, PCM 2018, Hefei, China, 18/9/21. https://doi.org/10.1007/978-3-030-00767-6_66
Zhao Y, Cheng X, Ikenaga T. Spatial pixels selection and inter-frame combined likelihood based observation for 60 fps 3d tracking of twelve volleyball players on gpu. In Cheng W-H, Yamasaki T, Ngo C-W, Hong R, Wang M, editors, Advances in Multimedia Information Processing – PCM 2018 - 19th Pacific-Rim Conference on Multimedia, 2018, Proceedings. Springer-Verlag. 2018. p. 716-726. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-00767-6_66
Zhao, Yiming ; Cheng, Xina ; Ikenaga, Takeshi. / Spatial pixels selection and inter-frame combined likelihood based observation for 60 fps 3d tracking of twelve volleyball players on gpu. Advances in Multimedia Information Processing – PCM 2018 - 19th Pacific-Rim Conference on Multimedia, 2018, Proceedings. editor / Wen-Huang Cheng ; Toshihiko Yamasaki ; Chong-Wah Ngo ; Richang Hong ; Meng Wang. Springer-Verlag, 2018. pp. 716-726 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{8d49762026bb43b3a544300e8743ff6a,
title = "Spatial pixels selection and inter-frame combined likelihood based observation for 60 fps 3d tracking of twelve volleyball players on gpu",
abstract = "3D players tracking plays an important role in sports analysis. Tracking of players contributes to high level game analysis such as tactic analysis and commercial applications such as TV contents. Many services like sports live and broadcasting have strict limitation on processing time, thus real-time implementation for 3D players tracking is necessary. This paper proposes a particle filter based 60 fps multi-view volleyball players tracking system on GPU platform. There are three proposals: body region constraint prediction, spatial pixels selection and inter-frame combined likelihood. The body region constraint prediction uses player’s body region as limitation in prediction to increase tracking accuracy. The spatial pixels selection method selects pixels for likelihood calculating to reduce calculation amount in spatial space. The inter-frame observation method does particle filter algorithm with two frames each time to reduce calculation amount in temporal space. Our experiments are based on videos of the Final and Semi-Final Game of 2014 Japan Inter High School Games of Men’s Volleyball in Tokyo Metropolitan Gymnasium. On the GPU device GeForce GTX 1080Ti, our tracking system achieves real-time on 60 fps videos and keeps the tracking accuracy higher than 97{\%}.",
keywords = "3D volleyball players tracking, GPU acceleration, Particle filter, Sports analysis",
author = "Yiming Zhao and Xina Cheng and Takeshi Ikenaga",
year = "2018",
month = "1",
day = "1",
doi = "10.1007/978-3-030-00767-6_66",
language = "English",
isbn = "9783030007669",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer-Verlag",
pages = "716--726",
editor = "Wen-Huang Cheng and Toshihiko Yamasaki and Chong-Wah Ngo and Richang Hong and Meng Wang",
booktitle = "Advances in Multimedia Information Processing – PCM 2018 - 19th Pacific-Rim Conference on Multimedia, 2018, Proceedings",

}

TY - GEN

T1 - Spatial pixels selection and inter-frame combined likelihood based observation for 60 fps 3d tracking of twelve volleyball players on gpu

AU - Zhao, Yiming

AU - Cheng, Xina

AU - Ikenaga, Takeshi

PY - 2018/1/1

Y1 - 2018/1/1

N2 - 3D players tracking plays an important role in sports analysis. Tracking of players contributes to high level game analysis such as tactic analysis and commercial applications such as TV contents. Many services like sports live and broadcasting have strict limitation on processing time, thus real-time implementation for 3D players tracking is necessary. This paper proposes a particle filter based 60 fps multi-view volleyball players tracking system on GPU platform. There are three proposals: body region constraint prediction, spatial pixels selection and inter-frame combined likelihood. The body region constraint prediction uses player’s body region as limitation in prediction to increase tracking accuracy. The spatial pixels selection method selects pixels for likelihood calculating to reduce calculation amount in spatial space. The inter-frame observation method does particle filter algorithm with two frames each time to reduce calculation amount in temporal space. Our experiments are based on videos of the Final and Semi-Final Game of 2014 Japan Inter High School Games of Men’s Volleyball in Tokyo Metropolitan Gymnasium. On the GPU device GeForce GTX 1080Ti, our tracking system achieves real-time on 60 fps videos and keeps the tracking accuracy higher than 97%.

AB - 3D players tracking plays an important role in sports analysis. Tracking of players contributes to high level game analysis such as tactic analysis and commercial applications such as TV contents. Many services like sports live and broadcasting have strict limitation on processing time, thus real-time implementation for 3D players tracking is necessary. This paper proposes a particle filter based 60 fps multi-view volleyball players tracking system on GPU platform. There are three proposals: body region constraint prediction, spatial pixels selection and inter-frame combined likelihood. The body region constraint prediction uses player’s body region as limitation in prediction to increase tracking accuracy. The spatial pixels selection method selects pixels for likelihood calculating to reduce calculation amount in spatial space. The inter-frame observation method does particle filter algorithm with two frames each time to reduce calculation amount in temporal space. Our experiments are based on videos of the Final and Semi-Final Game of 2014 Japan Inter High School Games of Men’s Volleyball in Tokyo Metropolitan Gymnasium. On the GPU device GeForce GTX 1080Ti, our tracking system achieves real-time on 60 fps videos and keeps the tracking accuracy higher than 97%.

KW - 3D volleyball players tracking

KW - GPU acceleration

KW - Particle filter

KW - Sports analysis

UR - http://www.scopus.com/inward/record.url?scp=85057227252&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85057227252&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-00767-6_66

DO - 10.1007/978-3-030-00767-6_66

M3 - Conference contribution

SN - 9783030007669

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 716

EP - 726

BT - Advances in Multimedia Information Processing – PCM 2018 - 19th Pacific-Rim Conference on Multimedia, 2018, Proceedings

A2 - Cheng, Wen-Huang

A2 - Yamasaki, Toshihiko

A2 - Ngo, Chong-Wah

A2 - Hong, Richang

A2 - Wang, Meng

PB - Springer-Verlag

ER -