Real-time Rendering Of 3D Complex Scenes

The field of real-time computer graphics is one of the fastest evolving fields within computer science and engineering. Ever increasing requirements on real-time and high degree of realism make 3D scenes rendering rather complex, and enhancing hardware speed along is far from enough for this issue. Therefore, algorithmic and software accelerating technologies have formed an important research field, which is, undoubtedly, very challenging for many years to come. In many computer graphics applications, such as network game, 3D navigation, flying simulation and virtual surgery, the following requirements seems to be mandatory: 1, users can interact with a virtual environment in real-time; 2, objects can be rendered in real time and updated dynamically according to the motion of the user viewpoint; and 3, the blink of graphics should be undetectable by human eyes.This dissertation focuses on real-time rendering of 3D complex scenes without sacrificing the degree of realism. Much effort was devoted to large area plant real-time rendering, water and light in complex environment and physics-based real-time cloth simulation.Existing L-System based plant growing models focus on the growing process of a singular plant, thus unsuitable for describing the interaction between plants and the environment, or among plants. Therefore we proposed object-oriented L-System, which is very powerful in describing complex and large scaled plant-environment interactions.For the large area plant real-time rendering issue, we focus on forest rendering and grassland rendering. For many years, large area forest rendering has been a very complicated research issue, which attracted applications, especially 3D games. In such applications, not only realism and real-time requirements, but also physics simulation and artificial intelligence are involved, thus making the problem much more challenging. We proposed a multi-layer large-scale forest-rendering model, in which scene segmentation, Slicing, MMI and Imposter management technologies are integrated. Using dynamic assignment under the new model, good balance of the realism and the timeliness was obtained. For large area grassland rendering, simple polygons are employed for the rendering of grass blades, and linear approach is employed to form grassland. Moreover, wind is introduced into the LBM model, which in turn simulates the grassland blew in wind. We also discussed the GPU-based accelerate rendering algorithm, which is very fast when the realism requirement is not quite rigid.Numerous issues should be addressed in complex nature environments, and this dissertation only focuses on water and light. For water surface simulations, GPGPU was employed to solve the two-dimension wave equation, and good simulation results were obtained. For the light model, deferred lighting and HDR technologies were analyzed, and a combination approach was proposed. Experimental results show that HDR-based deferred lighting technology can help to represent more detailed scenes. Even though this approach slows down the rendering speed, the latter is still above 25FPS.Cloth motion simulation was always a hot topic in computer graphics. To cope with requirement such as real-time, realism and stability, this dissertation focuses on enhancing both the model and the integral algorithm, and proposed a cloth movement simulation approach. It can help implementing cloth simulation systems with good performance in terms of speed and stability are easy to implement.