The Study Of Virtual Human’s Eye Movement And Facial Expression Synthesis

With the rapid development of computer technology and the technology of virtual reality, the study of face modeling and facial animation of three-dimensional virtual human has gradually become a hotspot and difficult problem in the research field of virtual reality. Now, there are methods in this study such as:the modeling method based on two-dimensional image, the modeling method based on3D scanners, the modeling method based on modeling software (such as3D Max, Maya and Poser) and so on. The researchers can really get more realistic face model by using the modeling method based on3D scanners than other modeling methods, but for the price of the3D scanners is too high now, this method does not have broad research significance. If using modeling software to model the face of human, it is difficult to get human face model of real life and the data of the model created by these modeling software is too large to easily control. Although many researchers research the modeling methods based on two-dimensional images in recent years, they model eye of human individually or model entire human face that lack of consideration about the detailed movement of the eye model. So it is the focus of my research that how to build widely studying, highly authenticity and easily controlling3D face model with details movement of human’s eye.For these problems, we present a face reconstruction approach on improved Candide-3model. Firstly we model eye in Candided-3model. And we specifically divide the eye modeling into eyeball modeling and eyelid modeling. For the modeling of eyeball, we use the combination of two hemispheres to model. For the modeling of eyelid, we use bezier curve to model the inner and outer edge of eyelid and the upper lid groove in Candide-3model. Then by using information of facial feature points and local least squares matching principle, we adjust the standard3D model to a special3D model. That special3D model is similar to the input facial photo. After that, we paste facial texture by feature point constraint method and paste texture to eyeball by partially spherical and spherical texture mapping method. Finally, in the built model, we realistically complete some actions such as the close, open and jump of eyelid and the rotation of eyeball and some related actions by the method of the piecewise linear interpolation.Finally, on the improved face model, we complete six expressions with detailed animations of human’s eye by combination of facial motion unit vectors according to the relationship between eye animations and emotional and the descriptions and definitions of six expressions in the MPEG-4standards. Experiments show that authenticity of improved real face model in my study is relatively high and animation of expressions on the model is easily to control. The improved real face model has wide applicability.