Groups Animated Real-time Rendering Technology

In recent years, crowd animations are becoming more popular in many fields. However, the traditional rendering methods on current hardware are far from enough for the task of rendering and animating crowds of virtual humans in real-time. Therefore, the technologies of the real-time rendering of crowd animations have formed an important research field, which is very challenging for many years to come. This dissertation focuses on real-time rendering of crowd animations to enhancing the speed of crowd rendering. Much effort was devoted to the CPU-based skinning algorithm for the characters in crowds, the real-time rendering algorithm of crowds based on LOD and impostors, and the Instancing method for GPU crowd rendering.The progress of the rendering of crowd animations is actually composed by many characters' rendering. For the rendering of single character, a good effect will be got by optimizing this operation on CPU when the application is GPU-restricted. This dissertation proposed a CPU-Based optimized algorithm to accelerate the rendering of matrix palette skinning by using the SSE(stream SIMD extend) instruction on Intel processor, which is very powerful in the rendering of crowd animations when the points of the character are weighted by many skeletons.There are a lot of researches focused on the real-time rendering technologies of 3D complex scenes, but there interests are almost oriented towards the static objects. For the real-time rendering of crowd animations, we proposed a novel hybrid rendering algorithm for crowd animations that solves the classic problem of degraded quality of image-based rendering methods and low speed of full geometry-based rendering. This is realized by using impostors for characters with low detail, while a full, geometry-based representation for characters with full detail. The result demonstrates this method capable of rendering crowd animations in real-time with a low rendering cost. With the development of the graphics processor, a new chance is coming for the real-time crowd animations. Because many characters in a crowd have the same mesh, this dissertation focuses on the reduction of the amount of draw calls, state change and the requirement of the memory, and proposed an instancing method for GPU crowd rendering by reusing the same mesh. Experimental results show that the Instancing-based crowd rending technology can help to enhance the speed of the rendering of crowd. Compared with the traditional crowd rendering method, the FPS of the rendering of Instancing-based method is 40% more quickly than the former's.