Motion Capture Assisted Animation: Texturing and Synthesis a Dissertation Submitted to the Department of Physics and the Committee on Graduate Studies of Stanford University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

This thesis discusses methods for using the information in motion capture data to assist in the creation of life-like animations. To address the problem of developing better methods for collecting motion capture data, we focus on video motion capture. A new factorization method is presented that allows one to solve for the model of the subject’s skeleton from a series of video images. No markers or special suits are required. The rest of this thesis discusses new techniques for flexible use of motion capture data after it has been collected. For the case of cyclic motions such as walking, we demonstrate a technique for complete synthesis. It begins with an analysis phase, in which the data is divided into features such as frequency bands and correlations among joint angles, and is represented with multidimensional kernel-based probability distributions. These distributions are then sampled in a synthesis phase, and optimized to yield the final animation. We also demonstrate methods for performing texturing and synthesis of widely varying motions based on motion capture data. First we present results using principle components analysis as a basis for the texturing and synthesis. Second we discuss our most successful technique for motion capture assisted animation, in which a simple matching algorithm is used. These methods allow an animator to sketch an animation by setting a small number of keyframes on a fraction of the possible degrees of freedom. Motion capture data is then used to enhance the animation. Detail is added to degrees of freedom that were keyframed, a process we call texturing. Degrees of freedom that were not keyframed are synthesized. The methods take advantage of the fact that joint motions of an articulated figure are often correlated, so that given an incomplete data set, the missing degrees of freedom can be predicted from those that are present. Finally, we discuss the various techniques and results, and suggest approaches for future improvements.