| Summary: | 碩士 === 國立中正大學 === 資訊工程所 === 96 === In most walkthrough systems, geometry data are stored on the local device. In such cases, the research usually aims at object culling for real-time rendering. However, with the increasing size of geometry dataset, the storage requirement far exceeds the capacity of a single local machine. It is becoming necessary to employ a large-scale centralized storage server for real-time support of several clients during visualization. Nevertheless, in such a distributed walkthrough system, network transmission is often the performance bottleneck. Therefore, methodology to select most significantly demanded objects under the limited bandwidth and limited cache size constraints is important. For solving the aforementioned problem, we develop an intelligent searching method for a client-server based walkthrough system, which helps clients in time to get the needed geometry objects in the virtual environment. Our work focuses on using a prioritized searching method to assign a priority value to each object within the search range. But there are several factors which have different measurement units that can influence the priority. It needs a scheme to convert those factors into the prioritized parameters. We consider the correlation of user state and object state, including their spatial and temporal information, to dynamically adjust the prioritized parameters of a Gaussian function. Gaussian function can convert the factor values of different parameters to the range we expected. Client therefore can get the display objects according to their prioritized order. Our method can also be integrated into a useful prefetching approach to a walkthrough system, to maximize the effectiveness of visualization, which utilizes device idle time to download or prefetch geometry data that the user might view in the near future. When applied the new developed search method to our walkthrough testbed, experimental results show a good performance increase during the visualization.
|
|---|
