Data-driven synthesis of realistic human motion using motion graphs

Realistic human motions is an essential part of diverse range of media, such as feature films, video games and virtual environments. Motion capture provides realistic human motion data using sensor technology. However, motion capture data is not flexible. This drawback limits the utility of motion capture in practice. In this thesis, we propose a two-stage approach that makes the motion captured data reusable to synthesize new motions in real-time via motion graphs. Starting from a dataset of various motions, we construct a motion graph of similar motion segments and calculate the parameters, such as blending parameters, needed in the second stage. In the second stage, we synthesize a new human motion in realtime, depending on the blending techniques selected. Three different blending techniques, namely linear blending, cubic blending and anticipation-based blending, are provided to the user. In addition, motion clip preference approach, which is applied to the motion search algorithm, enable users to control the motion clip types in the result motion.