Techniques For Data-driven Character Animation

With the rapid development of computer technology, motion capture equipment has entered the commercial and practical process. Therefore, three-dimensional computer animation, which is based on the motion capture, has become the mainstream of current research and the field of computer graphics. With the establishment of the large-scale three-dimensional human motion capture database, the researchers'focus has shifted to the analysis, management and reuse of the existing three-dimensional human motion capture data and the application of new technologies for automated 3D Human Animation.This thesis first systematically described the situation of current data-driven character animation's techniques, and then made detailed analysis and comparison on existing motion editing techniques and motion synthetic methods both at home and abroad in order to explain the advantages and disadvantages of various methods, focusing on the following issues:1. According to the definition of Quaternion and Euler angle, Quaternion's attributes. The conversion relationship of Euler angle and Quaternion are analyzed. and discusse the frame of capture data and Data-driven human animation in detail.2. According to the capture data, the original three-dimensional human motion capture data are segmented by the PCA segmentation algorithm and the Manifold Learning algorithm, then analyze and compare the advantages and disadvantages of each algorithm through the results of the experiment.3. We propose an algorithm of key frame extraction based the distance between two frames, encode chromosomes by a binary character, design the fitness function, then execute genetic operators. Finally, we find the optimal set of key-frame.4. A three-dimensional human motion synthesis experiment is designed based Behavior Finite-State Machine, construct motion states based on the key-frame. Motion states transit smoothly by the motion blending algorithm, and design an automatic movement generation module based Behavior Finite-State Machine.According to the oil-field maintenance-job stimulating operations, appoint the start and the end in the virtual scene, automatically plan the path and select the state based on Behavior Finite-State Machine, extract the corresponding motion clips from a motion capture database, transit the motion clips smoothly by the motion blending algorithm, and finally synthesize a motion sequence. Experimental results show a good visual effect.