Issues Of Modeling And Visualization Of Virtual Nature Scenes

Modeling and visualization of virtual nature scenes is a challenging research area, which is used in a wide variety of applications, such as GIS, 3D game, education, synthesized environment, flight simulation, digital exhibition and virtual battle.A nature scene consists of plants, terrain, sky, water, and various associated natural phenomena. Research on nature scenes is related to a wide variety of subjects including biology, physics, computer science and so on, therefore it is a typical cross-disciplinary research area. There exist a great number of ways to model and visualize nature sceness, yet most of them are complicated. Motivated by popular objectives from the view of computer science and technology, hoping to render large scale virtual nature scenes in a way of realisticity, real-time and scalibility, this paper addresses some new approaches to modeling and visualizing individual plants, rendering terrains in real time, and exhibiting large scale nature scenes on a display wall with high resolution and interaction functionality. Following are details about our research contents.Firstly, a new approach to model and visualize individual plants based on L-System is investigated. It usually is a process with intensive computation and datasets. This new approach takes the binary tree as a fundamental data structure, so that it could make great use of the internal branch structure of L-System. Based on this structure, a bracketed, stochastic, context-sensitive and parametric L-System is re-implemented. Compared with traditional L-System based on the array structure, the new algorithm has shown higher efficiency in terms of processing time and space, as well as scalability at the stage of plant development. Furthermore, this paper presents a new parallel algorithm for simulating the growth of individual plants, and also implements a parallel sort-last rendering algorithm. Results of numerical experiments have demonstrated that our parallel strategy works effectively on clusters with a multi-processors architecture.Secondly, a new chunked LOD algorithm based on Triangulated Irregular Netwrok (TIN) is proposed. With the rapid development of GPU technology, the focus of terrain rendering algorithms has been gradually moved from CPU to GPU, whose objective is to make use of the powerful capability of computation and rendering on GPU, so as to decrease the burden on CPU. Chunked LOD was designed originally for this purpose. However, the current version of chunked LOD is mainly aiming to semi-regular triangular network. The problem is that it easily produces redundant vertices, which will cause the grid scale tremendously expanded. Based on Bowyer-Watson incremental point insertion technique, this paper presents a robust Delaunay algorithm for TIN creation. The new algorithm avoids production of redundant vertices, therefore the grid scale has been greatly decreased about 3 times for typical applicagtion examples. The benefit is not only the decrease of interim file size, but also the efficiency of terrain rendering.Thirdly, a new approach to construct a semi-immersive display wall for exhibiting large scale nature scenes is introduced. It utilizes sort-first parallel rendering based on PC clusters for a high ratio of performance/price, a software seamless alignment method for great scalability, and also hand gestures based on computer vision for flexible human computer interaction. The major contribution includes: (1) A construction method of a display wall based on popular commodity hardware has been systematically studied, including system architecture, geometry calibration, color calibration, parallel rendering, software platform, and etc. (2) A hand gesture interaction method based on computer vision is introduced. The method detects hands by using the color segmentation model, tracks hands by the CamShift algorithm, extracts hand contour by Freeman algorithm, and at last captures hand gestures by fingers curvature. Once the hand gestures are recognized, they are then mapped to the common keyboard and mouse instructions, which are used to control roaming on nature scenes.To demonstrate the effectiveness of our research achievement, finally we have designed a platform for exhibiting complex nature scenes, which integrates both the algorithms of virtual plant modeling and terrain visualization mentioned above, as well as regular algorithms for visualizing other elements, including plant communities, water and sky. Several examples have been exhibited. Through hand gesture interaction, users are able to experience the dramatic feeling of interactive, half-immersive roaming on the beautiful nature scenes.