Skeletal similarity based automatic joint mapping for performance animation

Abstract

Motion capture technology is becoming increasingly readily available to end users. To accommodate this development we aim to create an application that can map an as wide array of characters as possible, using minimal assumptions. For the input multiple (non-)human characters can be used to animate a target character. By using only information that can be derived from the skeletons and no prior learning phase, the approach ensures flexibility. After selecting the input and target characters, users can get a decent mapping without changing any options, though if desired settings can be changed to influence the mapping. The mapping subdivides the skeletons into sections with characteristics and uses these when comparing all the bone matches. These comparisons are done using weighted metrics, which check various aspects to determine which matches have the best matching role and function in each skeleton. By taking the combined weight these metrics, the best matching input bone per target bone can be found.