Court-Divisional Team Motion and Player Performance Curve Based Automatic Game Strategy Data Acquisition for Volleyball Analysis

Xina CHENG  Takeshi IKENAGA  

Publication
IEICE TRANSACTIONS on Fundamentals of Electronics, Communications and Computer Sciences   Vol.E101-A   No.11   pp.1756-1765
Publication Date: 2018/11/01
Online ISSN: 1745-1337
DOI: 10.1587/transfun.E101.A.1756
Type of Manuscript: Special Section PAPER (Special Section on Smart Multimedia & Communication Systems)
Category: Systems and Control, Vision
Keyword: 
volleyball game analysis,  strategy data acquisition,  team status,  attack performance variables,  

Full Text: PDF(1.6MB)
>>Buy this Article


Summary: 
Automatic game strategy data acquisition is important for the realization of the professional strategy analysis systems by providing evaluation values such as the team status and the efficacy of plays. The key factor that influences the performance of the strategy data acquisition in volleyball game is the unknown player roles. Player role means the position with game meaning of each player in the team formation, such as the setter, attacker and blocker. The unknown player role makes individual player unreliable and loses the contribution of each player in the strategy analysis. This paper proposes a court-divisional team motion feature and a player performance curve to deal with the unknown player roles in strategy data acquisition. Firstly, the court-divisional team motion feature is proposed for the team tactical status detection. This feature reduces the influence of individual player information by summing up the ball relative motion density of all the players in divided court area, which corresponds to the different plays. Secondly, the player performance curves are proposed for the efficacy variables acquisition in attack play. The player roles candidates are detected by three features that represent the entire process of a player starting to rush (or jump) to the ball and hit the ball: the ball relative distance, ball approach motion and the attack motion feature. With the 3D ball trajectories and multiple players' positions tracked from multi-view volleyball game videos, the experimental detection rate of each team status (attack, defense-ready, offense-ready and offense status) are 75.2%, 84.2%, 79.7% and 81.6%. And for the attack efficacy variables acquisition, the average precision of the set zone, the number of available attackers, the attack tempo and the number of blockers are 100%, 100%, 97.8%, and 100%, which achieve 8.3% average improvement compared with manual acquisition.