基于多摄像头和一致性子集的乒乓球跟踪方法  

Table Tennis Tracking Method Based on Multiple Cameras and Consistency Subset

作　　者：李云[1] 侯力 刘立华 徐伟[1] LI Yun;HOU Li;LIU Li-hua;XU Wei(School of Physical Education and Health,Jiangxi University of Chinese Medicine,Nanchang 330004,China;Sports department,Renmin University of China,Beijing 100872,China;Sports department,Institute of Disaster Prevention,Sanhe 065201,China)

机构地区：[1]江西中医药大学体育健康学院,江西南昌330004 [2]中国人民大学体育部,北京100872 [3]防灾科技学院体育部,河北三河065201

出　　处：《控制工程》2022年第1期54-60,共7页Control Engineering of China

基　　金：江西省体育局课题项目(2015026);江西中医药大学课题项目(2018jzyb-28)。

摘　　要：乒乓球运动具有高速性,所以乒乓球跟踪必须满足低延迟和高采样率。为此,提出了一种基于最大一致性子集的乒乓球机器视觉跟踪方法。首先,使用启发式算法寻找目标像素的集合,确定乒乓球的位置;然后,为了纠正启发式算法可能的错误报告,移除离群点,对不同相机所报告的位置进行最大一致性检查,找到并丢弃在目标检测阶段中得到的错误位置;最后分别在仿真环境和现实机器人平台上对所提方法进行评价。与RTBlob等系统相比,所提方法的跟踪准确度更高,对离群点的鲁棒性更优。此外,随着相机数量的提升,所提方法的准确度和鲁棒性也会增加。Due to the high speed of table tennis,table tennis tracking must meet the requirements of low delay and high sampling rate.For this reason,a machine vision tracking method based on the maximum consistency subset is proposed.Firstly,the heuristic algorithm is used to find the set of target pixels and determine the position of table tennis.Then,in order to correct the possible error report of the heuristic algorithm,that is,to remove outliers,the maximum consistency check is carried out for the positions reported by different cameras,and the error positions obtained in the target detection stage are found and discarded.Finally,the proposed method is evaluated based on the simulation environment and the real robot platform.Compared with RTBlob and other systems,the proposed method has higher tracking accuracy and better robustness to outliers.In addition,as the increase of the number of cameras,the accuracy and robustness of the proposed method also increase.

关 键 词：乒乓球跟踪 最大一致性子集 离群点 启发式算法 鲁棒性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]