Real-Time Rendering And Dynamic Simulation Of Large-Scale Forest Scenes

Rendering large-scale scenes at interactive rates is an extremely challenging problem in three-dimensional computer graphics. Large-scale forest scenes are one of the most important natural scenes which are useful in virtual reality and digital entertainments. Rendering large-scale forest scenes in real-time is a great challenge since the storage costs and geometric complexity of these scenes are far beyond the capabilities of current graphics hardware.Current forest rendering and dynamical simulation algorithms are not capable to deal with for-est scenes with more than 100,000 trees. In order to reduce the storage costs and triangle number of forest scenes, this paper presents a GPU-based approach for real-time rendering and interactive simulation of large-scale forest scenes in wind. At first, we use an mixed approach to represent tree models in GPU-friendly way. Then we adopt a dynamic depth quad stream residing on GPU buffers to represent view-dependent forest scenes. At last, the natural wind is modeled as a sta-tionary stochastic process/field, and the motion of trees driven by wind force is simulated using a procedural animation approach. The result demonstrates that the proposed method is effective and practical for real-time rendering and fast simulation of large-scale dynamic forest scenes with up to one million trees at low storage costs, low GPU buffer costs and low CPU work load. Our contributions are as follows:·Propose a GPU-friendly tree model presentation. We simplify tree models into layered depth images and reconstruct depth meshes in different LOD levels.:The depth meshes in different sample directions are blended to present trees in a view-dependent way. The rendering effi-ciency of this approach is not relative to the original tree models, but depends on depth quad number. The approach is well-suited to GPU parallel pipelines and is useful for GPU-based rendering and simulation algorithms.·Propose an GPU-based algorithm for large-scale forest representation and rendering. We introduce a dynamic depth quad stream residing on GPU buffers to render view-dependent scenes based on optimized tree models. Depth quads in stream can be refined through a sequence of quad subdivision operations to yield detailed quad mesh. More importantly, these operations are well-suited to parallel processing in modern geometry shader, and this incremental mesh updating process requires no continuous data exchange between CPU and GPU. The results prove that the algorithm could render one million trees with 200 MB storage costs and 2,000 MB GPU buffer costs in real-time.·Propose an GPU-based algorithm for interactive simulation of large-scale dynamic forest scenes in wind. Since human eyes are not sensitive enough to notice small motion errors of large-scale forest scenes, we could a procedural animation approach instead of physically-based simulation. For this reason, we model the natural wind as a stationary stochastic pro-cess/field, and integrate the simulation procedures into our GPU-based rendering algorithms. The results show that the algorithm is able to interactively simulate large-scale forest scenes with insignificance costs.