Research On Visibility Culling Methods Based On GPU

In recent years, with the development of graphics, three-dimensional model instead of two-dimensional data to show the real world has been widely used in many fields for its intuition and reality. In the3D visual simulation system of high-speed railway, the scale of the scene is large in great majority, and the number of polygons sometimes reaches a million or more. High demands are put forward to the graphics hardware and rendering algorithm. Recently, graphics processor with high parallel GPU as the core is in a rapid development, and accelerating the velocity of graphics rendering is dependent on GPU, and reducing the interaction between CPU and GPU has become a hot topic. The paper performs research on how to improve the efficiency of GPU rendering. Good visibility culling algorithm can greatly reduce the burden on GPU. The main results of this paper are as follows:1. This thesis introduces a view frustum culling algorithm based on the compute shader:applying the new features of compute shader sending control commands to GPU and Single Program Multiple Data in the latest OpenGL standard to improve efficiency of view frustum culling. Compute shader do the view frustum culling, and at the same time the number of vertices to be rendered is recorded by atomic counter. View frustum culling based on the compute shader not only reduces the intervention of CPU to GPU rendering process, but also avoids from reading and writing between different buffers and processes data in batches with multiple threads. Therefore our algorithm gives full play of GPU.2. This paper performs a predictive GPU acceleration occlusion culling algorithm:a predictive judgment principle is introduced into existing all-instance occlusion culling calculation by the vertex shader and geometry shader. With the visibility information of recent period of time we can do the occlusion culling selectively, in order to reduce the burden on GPU as well as improving rendering efficiency. Sifting all the instances is according to the time continuity. When instances are forecasted as visible or invisible, they can be rendered or rejected without occlusion testing. Only when the visibility of the instances is uncertain, they should be sent to do the occlusion testing and culling. The conditional sifting effectively avoids doing occlusion culling with all instances every time. The prediction is done by the vertex shader, which does not need the intervening of CPU. The approach reduces the burden of GPU, and improves the efficiency of the occlusion culling.Above algorithms are applied to the terrain rendering. Experiment results show that the new features of OpenGL and predictive occlusion culling methods give full play to GPU performance and improve scene rendering rate.