Natural Human Motion Animation Based on a Natural Pose Hyperspace
My PhD research focuses on creating realistic human animation automatically. Human animation is either created by an artist (then interpolated using keyframing), or a motion is captured from a human actor. Both of these are quite cumbersome and are very limited in the number and naturalness of the motions created. To create motions automatically kinematics and dynamics methods can be used. The problem is that kinematics method do not necessarily create natural motions, while dynamics methods require the creation and use of complex models.
I attempt to create natural poses and motions automatically based on motion captured data. The characteristics of natural motion is first determined by analyzing the data from motion captured data. A hyperdimensional space of natural poses is then created. Motion is created by traversing the points in the natural pose space. The motion created are more natural than kinematics methods. Motion can be created by using existing motion captured data and expanding the data (no need to capture new data for each movement).
The algorithm proposed (leftmost human) creates more natural poses
The main hurdle in this research is finding a hyperdimensional space that encompasses natural poses. PCA, (Principal Component Analysis) a simple statistical method can transforms motion captured data to a natural space hyperspace. Enhancements to this natural space creation algorithm is being developed
The findings so far supports the claim that such a method creates more natural poses compared to kinematics methods. Although the created poses is still not as good as captured data.