Generating Motion Capture Data for Arbitrary Rigs

As motion capture techniques become more advanced and accessible, motion capture data has become an increasingly popular source of efficient, realistic animation. The process of motion capture involves placing markers on an actor’s body and recording movement data for each marker. This data can then be mapped onto a digital human skeleton, called a rig, which consists of joints (e.g. hips, elbows, wrists). Motion capture is a practical alternative to the more traditional method of hand-animating each joint in a rig. However, the data obtained by motion capture is highly specific to the rig used to record it, making it difficult to transfer recorded animation from one rig setup to another. There is a plethora of motion capture data available on-line, however 3D artists who already have an existing rig that is well suited for a unique character cannot use that animation. This project aims to break the dependence of motion capture on a specific rig by generalizing animation for arbitrary rigs using machine learning algorithms.