乒乓球运动员发球姿势三维图像校正仿真

摘要

对乒乓球运动员发球姿势三维图像的校正,能够有效提升乒乓球比赛质量.对三维图像的校正,需要将图像分解为基本分量与细节分量,对分离出的三维图像细节分量时域曲线进行梯度加权均值滤波,完成三维图像的校正.传统方法先设计理想基线和光心,估计乒乓球发球姿势三维图像细节深度信息,但忽略了对图像细节进行梯度加权均值滤波,导致校正精度偏低.提出基于三边滤波的乒乓球发球姿势三维图像校正方法.构造乒乓球发球姿势结构保持的三维图像运动伪影校正模型,该模型由图像非局部均值正则项和块相似保真项构成,再将乒乓球发球姿势三维图像分解为基本分量与细节分量,对三维图像细节分量的时域曲线进行梯度加权均值滤波.实验结果表明,所提方法能够较好地保留图像细节特征,且各项图像客观评价参数具有大幅度提升.%Traditional method of three-dimensional (3D) image correction of serving form of table tennis always neglects to carry out gradient weighted mean filtering for image detail,which leads to low correction precision.In order to solve the problem,based on trilateral filtering,a new correction method is proposed in this research.Correction model of motion artifact of the 3D image kept by structure of serving was built.The model consisted of regular term of non-local means and fidelity term of piece similarity.The 3D image was decomposed into elementary component and detail component.The gradient weighted mean filtering was carried out for time domain curve of the detail component.It is proved by experimental results that the method can reserve detail feature of the image preferably and improve various objective evaluation parameters by a large margin.