布料运动模型参数学习方法

Parameters Learning for Cloth Simulation

为了在布料运动仿真中通过调整其中的模型参数表现不同材质布料的运动差异,提出一种基于实例的布料运动模型参数学习方法.该方法利用运动捕获设备获取布料样本的运动实例数据;通过仿真数据与实例数据对比,利用遗传算法逐步进化、学习得到布料运动模型的全部参数.采用文中方法对6种不同材质布料的运动模型参数进行学习,并利用学习结果合成相应的布料运动模拟数据.实验结果表明,每段布料运动模拟数据(时间长度为5 s)与实例数据间的相对位置误差小于6%;在相同外力驱动下,不同材质布料的运动模拟结果能够较好地反映不同材质布料的运动特性.

In virtual reality applications,it＇s difficult to tune the parameters of the cloth model to present vivid features of different fabric materials.In this paper,a method of learning parameters from real data is proposed.First,real data of the fabric motion are captured by motion capture devices.Then,parameters of the motion model are optimized through a genetic algorithm aiming to minimize the differences between simulated data and real data.Using this method,we obtained parameters regarding 6 different kinds of fabrics.Experiments showed that,with parameters learnt from real data,the relative errors per 5 seconds between simulation data and real data are below 6%.Furthermore,under the same external force,the simulated data of different fabrics could diversely feature the corresponding real fabrics well.