服装动画中人体运动与服装变形的相关性学习

摘要

针对服装动画中人体运动与服装变形的关系首先对服装动画中人体运动特征和服装变形特征进行定义和提取;其次建立多套服装在多种人体运动下的高精度服装动画实例数据并基于4种机器学习模型从实例中学习二者之间的关系.对不同学习模型进行实验的结果表明由于服装动画中人体运动与服装变形之间存在较强的相关性因此利用人体运动可以较为准确地预测服装的变形分布;相比BP神经网络广义回归神经网络和支持向量机等方法随机森林模型可以更有效地获取人体运动与服装变形的关系最终误差可控制在理想范围内.%In this paper, we investigate the correction between human motion and garment deformation. First, the features of human motion and garment deformation were defined and extracted. Second, high-resolution garment animations were generated using different types of garment and human motions. The correction between them then was studied based on four machine learning algorithms. Experimental results show that the correlation between garment deformation and human motion is strong, and thus garment deformation distribution can be predicted based on human motion. Compared with BP neural network, generalized re-gression neural network and support vector machine, random forest method is more effective and more pre-cise, and the final error of random forest is in the ideal range.