Statistical analysis of natural human motion for animation

Generating human motion that appears natural is a long standing problem in character animation. Researchers have explored many different approaches including physics-based simulation, optimization, and data-driven methods such as motion graphs and motion interpolation. One major difficulty in applying most of these approaches is the lack of an implementable definition of what it means for motion to be natural or human-like. In this thesis, we explore two techniques to fill this gap. The first technique creates a naturalness measure for quantifying natural human motion. The second technique involves a statistical analysis of human motion to compute aggregate statistics that are needed to guide animation algorithms for human figures toward natural looking solutions. A naturalness measure should be useful in verifying that a motion editing operation has not destroyed the naturalness of a motion capture clip or that a synthetic motion transition is within the space of those seen in natural human motion. To develop such a measure, we argue that the evaluation of naturalness is not intrinsically a subjective criterion imposed by a human observer but is, instead, an objective measure that can be computed from a large set of representative motions. We base our approach on a statistical analysis of a large motion database. Using positive training data only, the system learns a set of statistical models that represent the motion of individual joints, limbs, and the whole body. Each model produces a score for the naturalness of the test motion and these scores are then combined into an aggregate score to classify the input motion as natural or unnatural. We present ROC curves of the performance of these techniques on a broad set of test sequences and compare the results to human performance in a user study. Aggregate statistics about the properties of human motion are needed to guide animation algorithms to generate natural looking motion. We compute and report a variety of statistics for joint angle range of motion, joint velocities, and dimensionality reduction using a large and representative motion capture database. We also develop new techniques for identifying motion synergies and summarizing motion in a visually intuitive way.