Modeling And Simulation Of Forest In Wind

The modeling and rendering of natural scenes is an important research topic, which plays a significant role in computer animation, military simulation, virtual reality, special effect for film and television, etc. Meanwhile, the dynamic nature scenes, especially the simulation of dynamic forest scenes has been a challenging task, mainly due to the high complexity of their structures and the large computational overhead of physics-based simulation.This dissertation firstly introduces the background and significance of forest's modeling and simulation in wind filed, summarizes the algorithm of trees'modeling, the technique for rendering and optimization of forest, and the simulation of forest in wind field. Then, in-depth research is carried out around forest's modeling and simulation in wind field, including: trees'modeling algorithm based on random point in specialized space, trees'physical animation in wind based on GPU, the strategies for improving forest's rendering and forest's simulation in wind. The main contributions of this dissertation are as follows:1) L-systems, IFS and fractal method are the prevailing modeling methods for trees, but their generated graphs are of strong self-similarity and unable to represent the differences of different trees. Based on the branching type and stochastic space points, a novel modeling method for trees are proposed. This method not only overcome the above shortcoming but also can easily control the braches'smooth level and density. Experiments show that the result graphs are true to nature and the modeling method can be applied to computer animation.2) The present methods used to simulate trees'physical animation in wind cost too much and do not fit for real-time occasions. This dissertation introduces a novel real-time tree animation method that is both significantly faster and more physically plausible than previous methods. Taking physical model of branches as linearly tapered circular beam, its bending effect is analyzed and the expression of its deflection curve is derived. To enhance computational efficiency, a least square fit is performed to the deflection curve equation and then its value is pre-computed to 2D-texture which is accessed by GPU run-time. To balance the load on CPU and GPU, we design a hierarchical tree structure model that transfer all the mechanical analysis and calculation to GPU, meanwhile, avoid to segment a branch to model its deformation behavior. Experiments show that the presented method can be completely real-time, realistically simulate the tree in wind, solving the problem of current method in literature that cannot simulate the physical real-time animation of tree in wind.3) To enhance the reality of forest, pre-calculated wind attenuation coefficient strategy is proposed and implemented after studying the principle of wind speed decay. According to the principle of wind decay, calculate the attenuation coefficient of every tree in different wind direction and then store it in special texture. In real-time simulation, sample every tree's attenuation coefficient to estimate the value of wind force which determines tree's movement amplitude.4) To reduce the cost of real-time computation of forest scene, the 2D-VFCA algorithm and texture look-up method are presented. 2D-VFCA completes frustum culling in 2D space unlike traditional view frustum culling in 3D space: Firstly, compute the eight points of view frustum in world coordination. Secondly, adopt improved Melkman algorithm to calculate the largest convex polygon of the projected area of view frustum in XOZ plane. Thirdly, judge whether the projected area of AABB bounding box of every object is in the largest convex polygon, if true, the object is in the view frustum and vice versa. Compared to traditional view frustum culling method, our presented method is four times faster and can be applied to real-time rendering of large scene. For complex multivariate function on GPU, texture look-up method is designed to enhance computational efficiency. Reconstruct the complex multivariate function to several two-variable functions and then map them to different 2D textures. When real-time running, one texture sampling result may act as texture coordinate for another texture and the final result of multivariate function is obtained by last texture sampling. Error analysis and texture setting are addressed in detail to make the proposed method easy to be implemented. Eliminating the high computing overhead of complex multivariate function, the real-time computation of texture look-up method is only concerned with sampling frequency. Consequently, the proposed method can be applied to computation-intensive scenarios. 5) On the basis of above research, prototype system of forest's simulation in wind is built, which verify the correctness and validity of research results.