Interactive Low-Dimensional Human Motion Synthesis by Combining Motion Models and PIK

This paper explores the issue of interactive low-dimensional human motion synthesis. We compare the performances of two motion models, i.e. Principal Components Analysis (PCA) or Probabilistic PCA (PPCA), for solving a constrained optimization problem within a low-dimensional latent space. We use PCA or PPCA as a first step of preprocessing to reduce the dimensionality of the database to make it tractable, and to encapsulate only the essential aspects of a specific motion pattern. Interactive user control is provided by formulating a low-dimensional optimization framework that uses a Prioritized Inverse Kinematics (PIK) strategy. The key insight of PIK is that the user can adjust a motion by adding constraints with different priorities. We demonstrate the robustness of our approach by synthesizing various styles of golf swing. This movement is challenging in the sense that it is highly coordinated and requires a great precision while moving with high speeds. Hence, any artifact is clearly noticeable in the solution movement. We simultaneously show results comparing local and global motion models regarding synthesis realism and performance. Finally, the quality of the synthesized animations is assessed by comparing our results against a per-frame PIK technique.