基于隐空间的运动捕获数据自动分割*

提出一种运动捕获数据自动分割方法。利用高斯过程隐变量模型对运动捕获数据进行降维，将其从高维观察空间映射到低维隐空间；在隐空间中构造运动特征函数，该函数具有结构简单、对所有关节敏感等优点，通过分析运动特征函数几何特征的变化，探测运动捕获数据的分割点，实现运动自动分割。实验结果表明，该方法具有较高的准确度和较好的普适性。

Automatic segmentation of motion capture data based on latent space

This paper proposed an automatic segmentation technique for motion capture data. It reduced the dimension of motion capture data with Gaussian process latent variable models, mapped the motion capture data from high-dimensional observation space to low-dimensional latent space. Construced motion character function in latent space, which had much excellence such as simple construction, sensitive to all joints, and so on. By analyzing geometry character of motion character function, it could detect the segmentation point in motion capture data, and segment the motion. Experiments show that this technique has high correct rate and well adaptation.