The Study Of Virtual Human’s Skeleton Extraction Based On Model Decomposition And Skeletal Animation

As the main body in the virtual world, virtual human carries the mission of replicating human behavior. And it has been a focus that how to make the avatar act lively according to men’s will. Now, skeletal animation has been used widely because it is simple and effective. But how to make the simple structure simulate the complex action of the human body, it is still a question that needs to be explored further more. There are important factors that influence the sense of reality such as the construction of 3D skin, accurate skeleton structure as well as the binding effect. In order to synthesize vivid and coherent motion of virtual human and achieve the lab goal of using text to drive the avatar to move and make expressions, skeletal animation based on skeleton extraction is studied deeply in this thesis.First of all, to meet the requirements of driving movements and expressions at the same time, the avatar’s body and head are made separately to construct the skin model. We use 3D modeling software to establish a human body without head from a 2D original painting, and connect the body with a head which is completed by the lab previous work to build a skin model which can satisfy the facial animation standard. But the synthesis of facial animation will be done by other members of our laboratory. Then, aiming at the questions that current skeleton extraction methods are complicated and few of them are specifically for human bodies, a method of extracting skeletons from 3D human models based on model decomposition is proposed. This method obtains the skeleton by analyzing both the branches and the topology structure of the model. The branches are calculated by the minmum geodesic distances between vertices and feature points and the topology is determined by the geodesic distance function as well as the partition threshold. In this thesis, the vertices belong to the same level are computed successively in the branches, and skeleton points can be close to the joints by discussing an appropriate threshold from the experiments. As a result, accurate skeletons with different postures are extracted from the models. Finally, after constructing the skeleton points hierarchically and calculating binding weights by the bone projection method, the skin of the model and the extracted skeleton are associated by linear blending skinning algorithm. Through the control of translations and rotations of the joint chain, a continuous and vivid animation of surface model is generated.The experimental results show that the method can get more accurate results with different postures and a realistic animation can be driven by the extracted skeleton.