REVIEW: a real-time virtual walkthrough system Lidan Shou, Jason Chionh, Kian-Lee Tan, Yixin Ruan, and Zhiyong Huang   View Paper (PDF)   Return to Demostrations Abstract 1. Real-time Virtual Walkthrough Data representing virtual environments (VEs) are getting increasingly large in order to better simulate real scenes. This poses interesting challenges to organize, store, and render the data for interactive navigation in VEs, or walkthrough. A large VE usually consists of thousands of 3D objects, each of which can be represented by hundreds of polygons, and may take thousands of megabytes of storage space. The amount of data is so large that it is impossible to store all of them in the main memory. Even for memory resident models, the graphics pipeline can become a bottleneck quickly with a large amount of data and slow down the rendering to an unacceptable frame rate for the walkthrough. In this demo, we will show the effectiveness of several optimization techniques to address the problems. (1) The data in secondary storage are organized based on their spatial location in an R-tree index and only relevant data are retrieved from the database using a novel R-tree search algorithm. (2) Prefetching of the R-tree index nodes and data is implemented based on a realtime prediction algorithm. (3) A cache replacement policy is used based on the hierarchical structure of the R-tree index and the walkthrough semantics. DiSC'02 © 2003 Association for Computing Machinery