Research On The Key Technologies In Simulation Of Large Scale Assembly Scene

Simulation of large assembly scale scenes is a very important topic in computer graph and virtual reality. With the wide application of digital design and digital plant, high efficiency rendering method is becoming more and more important. Moreover, high efficiency rendering engine is also required in many aspects such as virtual combat systems in military, virtual manufacturing systems in industry and3D animation games. However, the increasing data size of scene brings many difficulties for further application. To solve the problems caused by large scale scene date, this paper discusses these aspects as follows:To reduce the scene data, we propose a method to simplify the geometric model. Utilizing the secondary development of commercial modeling software, the hierarchy tree of assembly is traversed. During the process of traversal, the position of each part model is computed and saved in simulation database. At the same time, the part model is also exported into mesh model. When the traversal is completed, all the solid models are converted into mesh models. After that, some sub-assembles based on mesh model are selected for rendering. By rendering of the selected sub-assemble from many viewpoints, the rendering results are analyzed. We could conclude the invisible parts and facets. For the invisible parts, we could delete them from sub-assemble directly. For the invisible facets, the Edge Collapse method is used to reduce the invisible facets.To reduce the amount of calculation for collision detection, we design the flow of collision detection based on Hierarchical Bounding Volumes (BVHs). As the BVHs for each dynamic object should be rebuilt each frame, the BVHs building method is the key issue to enhance the efficiency of collision detection. For all the BVH building methods, the only different is how to choose the split plane for each node in the BVH. We study the theory which builds the optimal BVH in depth and propose an algorithm to probe the sub-interval which holds the optimal split plane. This can greatly reduce the split plane candidates and reduce the BVH construction time. To avoid the rendering system into a temporary interruption while the BVH is rebuilding, we use multi-thread to deal with this problem. One thread is used for rendering, the other for BVH rebuild asynchronous. During the period for the start of BVH rebuild to the completion of the next BVH, two BVH refit operations proposed in our paper are used to enhance the quality of BVH. The experimental data shows that the BVH rebuild asynchronous method improves the stability of frame rate.To generate more realistic image by rendering, we propose an algorithm for ray tracing in CUDA (Compute Unified Device Architecture). In ray tracing algorithm,95percent time is spent on calculating the intersection between a ray and a triangular facet. The main problem to design algorithm used for calculating the intersection between a ray and a triangular facet is to traversal BVH stackless. In this paper, by adding two properties in the node of BVH, that is, the pointer to the parent node and the current node type, stackless BVH traversal is realized. The other problem for program executed in CUDA is the long time-delay to fetch data from global memory. In this paper we use the texture memory to save the geometry information and BVH structure data. To avoid the problem caused by buffer pointer, we resave the BVH by array and the point is replaced by the index of array.To enhance the frame rate of rendering of large scale scene, we design parallel rendering based on deferred shading. Existing parallel rendering system can not deal with rendering with complex light calculation. Moreover, when rendering large scale scene, there are many invisible facets in each frame, and deferred shading can avoid rendering of those invisible primitives. Therefore it can greatly enhance the rendering efficiency. The deferred shading is achieved in two phases. The first phase is to render the primitives without shading. The second phase is to render the visible primitives. The load balancing of the first phase can be achieved by sub-dividing the primitives evenly. The load balancing of the second phase is achieved by sub-dividing the colored pixels evenly. The experimental result shows parallel rendering based on deferred shading is suitable for rendering of large scale scene. Compared with Sort-Last, it can obtain better load balancing and higher frame rate.To meet the requirement of project—Digital Simulation of High Power Laser Device, we develop a prototype system for rendering of large scale scene. By virtual assembly simulation, many issues in practice have been discovered. The algorithms proposed in this paper are also validated by this prototype system.