Octree-Based Optimization for Large CAD Models in VR : Rendering Efficiency Under Resource Constraints

Sammanfattning

In world of complex and increasingly massive three-dimensional datasets, a demand for efficient techniques for storage and rendering has emerged. The data structure octree is a popular solution within this field, providing an efficient structure for managing the large amounts of information. One such three-dimensional dataset is large 3D triangle meshes, commonly found in complex CAD models. The purpose of this thesis was to identify advantages and obstacles of using octrees to partition and visualize complex CAD models, specifically in the resource-constrained environment of a VR device. A VR application was developed where a large CAD model's 3D triangle mesh underwent partitioning into an octree. The octree model was then further equipped with a Level of Detail system to adjust the mesh's detail level based on distance from the camera, enabling the total triangle count to be decreased while keeping its essential detail up close. View frustum culling was implemented to remove unnecessary information from the rendering process, culling any parts of the model outside of the field of view. These techniques along with an Out-of-Core system for storage at the local disk allowed freeing up memory otherwise used by mesh information that was not to be displayed at that moment. Testing the application and tracking its performance metrics, it was shown that the performance of the program was significantly increased, as frame rate was consistently higher for the octree model than for the original model. Some advantages that were identified with the octree structure was the ability to treat parts of the model as separate entities, which allowed for any specific part to effectively be culled or have its mesh complexity adjusted, and free up memory for any information not currently needed. Some obstacles with this approach included a lengthy construction time of the octree largely occupied by mesh simplification, as well as occasional performance issues relating to frequent reading and writing large amounts of data to the disk.