Research And Implementation Of Real-time Rendering Technology, The Role Of The Large-scale Mass .3 D Scene

In recent years, crowd animation has been widely applied in many fields, including digital entertainment, training systems, urban planning, artificial intelligence research and so on. Therefore crowd real-time rendering, as an important component of crowd animation, has become a study object of many experts and scholars. At present, the studies mainly focus on two major aspects: one is the scale of the virtual humans of crowd, and the other is the reality of the appearance of the characters in crowd. However, the quantity and realism of the virtual humans in the crowd are mutual contradiction. Because if temporarily unable to improve the performance levels of the hardware and the technique of crowd real-time rendering, increasing the realism of the role is bound to increase computation workload, thus which will reduce the number of the role. Conversely, only reducing the reality of the appearance of a single role can increase the number of the character in the crowd. So based on the real-time requirements, it is very challenging to increase the number of character and enhance the realism of the visual appearance of virtual human.In order to increase the number of roles in the crowd, we can use a variety of acceleration technologies of large-scale crowd, such as the geometric model simplification, culling algorithm, instancing technology and so on. The thesis proposed a multi-resolution geometric modeling technique based on the study of the traditional geometric modeling techniques and model simplification. The technology introduces a multi-level LOD concept. It divides all the virtual humans of the whole scene into three rough LOD levels according to the distance, respectively using impostor, point model and the polygon mesh model. Then it divides the polygon mesh into three smaller LOD to form an approximate continuous LOD modeling. In addition, this thesis also proposed a method that can pre-process and render point model based on the DirectX 10 geometry shader, and can improve the efficiency of point model rendering to a certain extent by using geometry shaders.The research on the realism of appearance of the character in a crowd mainly focused on the crowd appearance variety and lighting calculations. In the large-scale crowd, if all of the virtual humans'appearance were exactly the same, then no matter how large the rendering number is, it can only feel like simple copy with no realism at all. Therefore, the thesis discussed the crowd appearance variety respectively from the role modeling, shape and texture appearance. And we proposed a mask mesh technique, which can generate the appearance for more different roles according to few mesh templates. On the shape change of a virtual human, we also proposed program logic based on skeletal deformation technique, the method can be used to change transformation matrix of the skeleton of the skinned mesh model dynamically during run time of the program, which can change the appearance of the character by controlling the matrix.The thesis also designed a crowd animation rendering system to verify the theory mentioned above. Tens of thousands of high-quality virtual characters can be real-time rendered in experiments, and the rendering system has applied in several simulation projects, for the traffic simulation, theater fire simulation and so on.