Abstract:

An inverse kinematics solution based on statistic learning is presented. Because of the high -dimension of character animation motion data and correlation lying in various dimensions, it is a very hard work to analyze and compute directly. The motion data are mapped from high-dimensional space to two-dimensional latent space based on Gaussian process latent variable models (GPLVM). Then, the representative poses of virtual characters are found out by clustering the motion data in latent space, which can expand a subspace that contains the primary characters and disciplinarians of training data. Finally, the weight of representative poses is optimized combined with constraints on the end effectors, and the optimized pose is obtained. The experiments show that the proposed method obtains a better effect.