Body part connection, categorization and occlusion based tracking with correction by temporal positions for volleyball spike height analysis

Xina CHENG*, Ziken LI, Songlin DU, Takeshi IKENAGA

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

3 Citations (Scopus)


The spike height of volleyball players is important in volleyball analysis as the quantitative criteria to evaluation players' motions, which not only provides rich information to audiences in live broadcast of sports events but also makes contribution to evaluate and improve the performance of players in strategy analysis and players training. In the volleyball game scene, the high similarity between hands, the deformation and the occlusion are three main problems that influence the acquisition performance of spike height. To solve these problems, this paper proposes a body part connection, categorization and occlusion based observation model and a temporal position based correction method. Firstly, skin pixel filter based connection detection solves the problem of high similarity between hands by judging whether a hand is connected to the spike player. Secondly, the body part categorization based observation uses the probability distribution map of hand to determine the category of each body part to solve the deformation problem. Thirdly, the occlusion part detection based observation eliminates the influence of the views with occluded body part by detecting the occluded views with a trained classifier of body part. At last, the temporal position based result correction combines the estimated results, which refers the historical positions, and the posterior result to obtain an optimal result by degree of confidence. The experiments are based on the videos of final and semi-final games of 2014 Japan Inter High School Men's Volleyball in Tokyo Metropolitan Gymnasium, which includes 196 spike sequences of 4 teams. The experiment results of proposed methods are that: 93.37% of test sequences can be successfully detected the spike height, and in which the average error of spike height is 5.96 cm.

Original languageEnglish
Pages (from-to)1503-1511
Number of pages9
JournalIEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences
Issue number12
Publication statusPublished - 2020 Dec


  • Body part categorization
  • Body part connection
  • Result correction
  • Spike height
  • Volleyball analysis

ASJC Scopus subject areas

  • Signal Processing
  • Computer Graphics and Computer-Aided Design
  • Electrical and Electronic Engineering
  • Applied Mathematics


Dive into the research topics of 'Body part connection, categorization and occlusion based tracking with correction by temporal positions for volleyball spike height analysis'. Together they form a unique fingerprint.

Cite this