| Simulation of cloud scenes becomes a key element to argument reality of natural environment in many applications such as virtual battlefield, virtual geographical environment, movie animation, 3D games and meteorological analysis. Along with the development of rendering techniques in virtual scene, the simulation and representation of clouds has become one of the hotspots in computer graphics and virtual reality in recent years. The realistic simulation of clouds not only can raise the fidelity of virtual scenes, but also can deliver abundant weather information. The simulation can help users understand and accept the weather information more intuitively. It also supports decision-making of any kind of weather-depended activities.However, most applications use cloudmap which provide a 2D view of clouds to represent meteorological cloud data, thus barely satisfy the visual requirement when the view point is near to clouds. While other researches of cloud visualization in computer graphics or some other specific application fields mainly focus on the detailed algorithms of modeling and rendering, which refer to quite limited data. So the research on realistic simulation of clouds based on meteorological data needs to be further discussed.The research of this thesis mainly focuses on the realistic simulation of cloud scene based on multi-scale meteorological data. The content includes topics like meteorological data based cloud modeling, organization of data and effective rendering techniques. Some corresponding novel algorithms and methods are proposed during the process of research and practice. In detail, the main contributions of this thesis can be concluded as follows:A sketch based hierarchical procedural implicit modeling method is proposed. Aming at dealing with meteorological data observed from ground, based on the sketch which user input, a sketch inflation method is designed to interpret the sketch. Meanwhile, a metaball fitting algorithm is designed to match the expanded field which results in a macro structure of clouds. Multi-scale noises are added to create detail level of clouds, and the proposed detail enhancing technique preserves the original macro-structure. An automated evolving particle system is also designed to generate metaballs which describe the detail of clouds. The models created with these methods not only match the basic shape defined by the sketch, but also have the characteristics of cloud shape. Furthermore, a hardware accelerated volume rendering algorithm is raised to render the modeling result at real time, providng an interactive modeling ability.A metaball fitting algorithm is proposed to generate cloud fields that match the given large-scale meteorological data, parameters of the metaballs are optimized by error control method. Also, a wavelet noise based high resolution detail generation method is proposed, in which the high resolution details are then merged into the original data field, the details are consistent with the original data in structure. The hierarchical organization method of multi-level cloud data is studied. On this basis, squadtree and octree are used to organize the volumetric data field by using structure related division of cloud data, together with wavelet analyse, which is profitable for rendering of large-sacle data.The basic idea of using potential field to control cloud animation is explained based on the equation of clouds'motion and then the specific design method of potential field and corresponding calculation method are proposed. Another control method using attraction forces in the boundary is also proposed. According to the equations of thermodynamics, a method which affects the motion of clouds by imposing the control of phase transition and latent heat is represented. Using these control operation, so that the clouds'motion is close to the target shape defined area, the motion obeys the basic motion rule of clouds.A discrete method of illumination model is discussed, and a multi-pass algorithm of illumination is proposed in order to balance the calculation task, and alleviate the rendering burden. The characteristic of the scattering phase function is analyzed, and then an illumination texture based acceleration method is proposed. The schedule scheme, acceleration scheme and caching scheme are given to handle large scale data and accelerate rendering. A rendering strategy combined several rendering techniques is designed based on the structural property of cloud data and vision parameters, which both fulfills the resolution requirement and avoids rendering of unnecessary data. Thus, the method not only accelerates rendering but also achieves interactive performance.A prototype system that represents cloud scene in virtual battlefield is implemented,. The methods about concerning on realistic simulation of cloud scene based on meteorological data proposed in this thesis are validated in this system, which has been applied in correlative projects. |
