针对现有的回收机器人对羽毛球的跟踪预测精度低问题,基于双目视觉技术,提出一种奇异值分解(Singular Value Decomposition,SVD)与混合高斯模型(Gaussian Mixed Model,GMM)结合KEF的羽毛球跟踪方法.该方法首先利用双目视觉技术采集羽毛球运动图像;然后通过SVD-GMM对羽毛球图像特征进行提取,并进行羽毛球三维轨迹重建;最后基于扩展卡尔曼滤波(Extended Kalman Filter,EKF)对羽毛球轨迹进行跟踪与预测.结果表明,SVD-GMM模型的羽毛球目标检测错检率、漏检率分别为0.5％、1.00％,羽毛球占比的平均误差值仅为0.14,单次羽毛球检测的平均耗时为0.86 s,具有较高的羽毛球提取精度和效率;基于EKF的跟踪预测模型输出的羽毛球落地位置的准确度较高,求解出的羽毛球落地位置较为精准;基于双目视觉技术的羽毛球跟踪预测方法能够实现对羽毛球落地时间及位置的精准预测,在羽毛球自动回收问题中具有一定的应用价值.
Target tracking and prediction method for badminton recycling robot based on binocular vision
Aiming at the problem of low tracking and prediction accuracy of existing recycling robots for badminton,a badminton tracking method based on binocular vision technology is proposed,which combines Singular Value Decomposition(SVD)and Gaussi-an Mixed Model(GMM)with KEF.This method first utilizes binocular vision technology to capture badminton motion images;Then,SVD-GMM is used to extract the features of the badminton image and reconstruct the three-dimensional trajectory of the badminton;Finally,based on the Extended Kalman Filter(EKF),the badminton trajectory is tracked and predicted.The results show that the SVD-GMM model has a badminton target detection error rate of 0.5％and a missed detection rate of 1.00％,respectively.The aver-age error value of the badminton proportion is only 0.14,and the average time for a single badminton detection is 0.86 seconds.It has high badminton extraction accuracy and efficiency;The tracking and prediction model based on EKF has a high accuracy in out-putting the landing position of badminton,and the calculated landing position of badminton is relatively accurate;The badminton tracking and prediction method based on binocular vision technology can achieve accurate prediction of the landing time and position of the badminton,and has certain application value in the automatic retrieval problem of badminton.
singular value decompositionmixed gaussian modelkalman filter algorithmbhattacharyya distance
万方数据
维普
