Converting Motion Capture Into Editable Keyframe Animation: Fast, Optimal, and Generic Keyframe Selection

Motion capture is attractive to visual effects studios because it offers a fast and automatic way to create animation directly from actors' movements. Despite extensive research efforts toward motion capture processing and motion editing, animations created using motion capture are notoriously difficult to edit. We investigate this problem and develop a technique to reverse engineer editable keyframe animation from motion capture.  Our technique for converting motion capture into editable animation is to select keyframes from the motion capture that correspond to those an animator might have used to create the motion from scratch. As the first contribution presented by this thesis, we survey both traditional and contemporary animation practice to define the types of keyframes created by animators following conventional animation practices. As the second contribution, we develop a new keyframe selection algorithm that uses a generic objective function; using different implementations, we can define different criteria to which keyframes are selected. After presenting the algorithm, we return to the problem of converting motion capture into editable animation and design three implementations of the objective function that can be used together to select animator-like keyframes. Finally, as a minor contribution to conclude the thesis, we present a simple interpolation algorithm that can be used to construct a new animation from only the selected keyframes.  In contrast to previous research in the topic of keyframe selection, our technique is novel in that we have designed it to provide selections of keyframes that are similar in structure to those used by animators following conventional practices. Consequently, both animators and motion editors can adjust the resulting animation in much the same way as their own, manually created, content. Furthermore, our technique offers an optimal guarantee paired with fast performance for practical editing situations, which has not yet been achieved in previous research. In conclusion, the contributions of this thesis advance the state of the art in the topic by introducing the first fast, optimal, and generic keyframe selection algorithm. Ultimately, our technique is not only well suited to the problem of recovering editable animation from motion capture, but can also be used to select keyframes for other purposes - such as compression or pattern identification - provided that an appropriate implementation of the objective function can be imagined and employed.