Parallel And Real-time Rendering Of Streaming Meshes

With the development of the 3D scanning and modeling techniques, three dimensional models are more precision and larger scale. The fact brings great pressure to geometry processing ,e.g., rendering, transmitting, and editing operations, since traditional digital geometry processing methods must load the whole original model into main memory to process. However, the capacity of main memory is limited, so traditional methods are unable to handle large-scale models that near or over memory capacity. Therefore, researchers have to seek geometry processing methods based on external memory, known as out-of-core algorithms. Now, the out-of-core algorithms become the hot of large-scale models processing.Out-of-core algorithms usually re-organize the data set of a 3D model on external memory in some ways, and access and load the local data sets into main memory during computing. However, most out-of-core processing algorithms have limits, e.g., the pre-process of re-organizing the data set is commonly time-consuming and the underlying data structure could be very complex and inconvenient to implement multi-resolution modeling. Solve these problems, streaming mesh, a new format of large models, is proposed to render large complicated mesh models conveniently.In this paper, we used streaming meshes and parallel processing methods to implement real-time rendering of large scale mesh.(1) Firstly, a parallel rendering method of streaming meshes was designed. A thread was created for each mesh, and more than one model was rendered in by mutex semaphore. Result showed that this method could enhance the multi-mesh rendering.(2) Secondly, an effective multi-resolution for large-scale mesh models was built and level-of-detail (LOD) was implemented. Dynamic control of LODs of models were implemented according to the distance of viewers' gaze point to each mesh. So, smaller data, which can meet users' visual demand(3) Thirdly, large-scale mesh was segmented into sub-parts by some parallel planes. For all sub-meshes, the former parallel rendering method was employed to accelerated the visualization; More important, the large-scale mesh could be totally rendered in one iteration, since a non-trivial re-construction method was developed by using of corresponding boundaries among parts. Besides these, level-of-detail of each part was built.