Research On Semantic Multi-Characters Deformable Motion Model

Crowds of people can be seen everywhere in real life. Real crowd motion can make the virtual environment seem realistic. With the rapid development of computer vision and computer graphics technology, more and more scholars pay attention to crowd motion simulation, which has become a research hotspot. Motion synthesis technology based on motion capture data can reduce cost of animation creation and improve the efficiency of synthesis. However, it needs massive calculation and storage space, and also needs to consider the interaction among the characters for simulation large-scale crowd motion. How to synthesis abundant, realistic and controllable crowd animation has important significance and applications.With the problem of high dimension and poor controllability of crowd motion data, this thesis decompose and reduce dimension of the data, and construct a deformable motion model for multiple characters. We use the deformable motion model for semantic multi-characters motion analysis and synthesis, combined with motion patches and motion graph. The main contributions of this thesis are as follows:1. A deformable motion model for multiple characters is proposed, which decomposes the crowd motion data into geometry and time variations. Extract keyframes and do DTW to the raw data, use PCA to reduce dimension, and then construct a low-dimensional semantic space of multi-characters deformable motion model.2. Propose a method of combining multi-characters deformable motion model with multi-characters motion patches. Construct multi-characters interactive events as motion patches, and tile the patches by random sampling and deterministic searching. Synthesize interactive multi-characters motion by achieving the mosaic continuously stitched by patches.3. Propose a two-layer structure of deformable semantic motion graph, which is able to synthesize different types of motion and different styles of motion of the same type.