面向类人角色动画的骨骼运动数据生成算法

为降低获取与给定类人角色模型相匹配的骨骼运动数据的成本,提高运动数据的可重用性,文章提出一种仅使用较少种类的动作捕捉数据获取多种类角色模型的骨骼运动数据生成算法。该算法首先利用动作捕捉数据训练运动数据自编码器,然后在该自编码器生成的隐变量空间中通过迭代约束求解的方式获得多种不同类人角色模型的骨骼运动数据。在 Adobe Mixamo 数据库上的实验表明,该文提出的算法具有较好的表现性能,能够有效生成多种类的骨骼运动数据,可以作为一种扩展骨骼运动数据库的方法,能够为影视、动画和游戏等领域降低动画制作的成本。

In order to reduce the cost of obtaining skeletal motion data matching a given humanoid character and improve the reusability of the skeletal motion data, this paper proposes an algorithm for generating skeletal motion data using only a few types of motion capture data to obtain multiple types of humanoid characters. The algorithm trains an autoencoder using a small amount of motion capture data, and then obtains skeletal motion data for many different humanoid characters by iterative solving in the latent space generated by autoencoder. The results of experiments on the Adobe Mixamo database show that the algorithm proposed in the paper has better performance and can effectively generate various types of skeletal motion data. It can be used as a method to expand the skeletal motion database and reduce the cost of movies, animations, games and other fields.