Highly adaptable real-time Kinematic Chain-Based Motion Retargeting for Virtual Characters

Abstract

Transferring motion from one character to another is a crucial operation in character animation. This procedure, also known as motion retargeting, is essential for many applications, including computer-generated films, virtual reality and video games. We present a novel kinematic chain-based approach for real-time animation transfer between virtual characters of diverse skeletal structures and proportions. Our method automates key setup processes, including armature matching by extracting non-branching joint sequences, and employs a novel heuristic skeleton matching to probabilistically establish limb correspondences based on joint positions and chain alignment. For motion transfer, a joint angle imitation technique intelligently adapts source animation to the target, even with rest pose differences. To reduce artifacts like foot skating or self-penetration we use a dynamic chain scaling method that adjusts target positions based on limb length ratios. Our framework is characterized by its high adaptability and extensibility, allowing distinct inverse kinematics solvers for individual limbs, providing a flexible open-source framework for animators and researchers.