In computer animation, virtual humans' realistic body and hand motions can be achieved through a combination of motion capture and 3D rendering software. However, the problem of perceptually incorrect finger and hand motions still plagues animators and researchers alike. My primary goal is to find ways of improving errors in data-driven approaches of synthesizing finger motions, by evaluating dis- tance metrics by human perception and analyzing human perception of finger poses.

This thesis examines the role of distance metrics for hand poses and finger joints as complimentary to body and hand-body relational metrics. More specifically, this thesis provides approaches to finding distance metrics which may be used in machine learning. I present algorithms for extracting rotation data from finger joints of 3D figures in multi-frame (animated) and single frame (static) poses. I also present a ranking system for similarity of static poses and similarity of animated clips.

Ultimately I discuss the use of distance metrics of fingers as valid data and show that they can be used to distinguish differences in poses. The limitations of distance metrics are explored in the human perception experiment. This thesis adds to the body of work using joint rotation distances. Additionally, this thesis contributes to human perception experiments of virtual characters, and extends existing databases with additional finger pose models.