Efficient simulation of large-scale urban environment with adaptive multiple-scene management

• 主要作者: Le, Minh Duc
• 其他作者: Seah, Hock Soon
• 格式: Theses and Dissertations
• 語言: English
• 出版: 2014
• 主題: DRNTU::Engineering::Computer science and engineering
• 在線閱讀: http://hdl.handle.net/10356/61564
• 機構: Nanyang Technological University

Project sizes and complexities of 3D Games and Simulations have increased drastically during the last ten years. Nowadays, a large number of applications have increasing scene complexity with non-scene-boundaries and contain many types of objects, both outdoors and indoors. There is, however, no general scene management system that can efficiently handle all scene types. It is already a heavy burden to deal with each scene individually. It definitely becomes very challenging to handle the combination of multiple-scene types. Another challenge is the increasing complexity of Physics and Artificial Intelligence (AI) simulation. Optimization becomes harder especially when there are a lot of agents involved and when resources are limited. There are also too many high-level algorithmic choices that require years of experience to master and identify the most suitable algorithms for specific tasks. We identify these challenges as our main research thrusts. Our approach is to research on high-quality scene management techniques to enhance real-time rendering performances of graphics pipeline and better support for handling AI and dynamic objects. As a core design concept, we decouple the scene graph structure from its content and flattening the inheritance structure of the scene graph and its data nodes. This provides an elegant and powerful architecture that allows for greater accommodation for various scene types as well as easy implementation of alternate data types. As the main research contributions, we proposed an Adaptive Multiple-Scene Management Framework (AMS) that can seamlessly handle all scene types as individuals or as a combination. We also identified a logical linkage between scene management concept and the hierarchical structures within the methodologies to handle AI and dynamic objects. We conceptualized this linkage at a higher level: using our proposed multiple-scene management framework to optimize AI and Physics activities such as pathfinding, decision-making, formation, collision detection and handling large number of dynamic objects. Keywords: scene graph, scene management, level of detail, culling, AI, dynamic object