Summary: | 碩士 === 國立成功大學 === 資訊工程研究所 === 88 === Recently, image-based rendering (IBR) has become one of most active research areas in computer graphics. This technique is a hybrid of both computer graphics and computer vision. Compared with traditional graphics, IBR has two advantages: (1) its complexity is in proportion to image resolution rather than scene complexity, and (2) it potentially saves time-consuming modeling time to generate highly realistic objects.
In this thesis, we employ the image-based rendering technique to walkthrough complex scenes that usually consist more than one million polygons. First, we build a hierarchical bounding volume tree. Second, we find a camera-sampling sphere for each node in this tree. Then, we optimally find a set of reference images for each sampling sphere. Finally, to walkthrough a scene, we adaptively traverse this tree and render each node in either warping reference images or drawing polygons.
We experimentally evaluate the proposed camera sampling method and walkthrough scenes. As a result, the proposed method performs much faster than traditional graphics rendering method. The reported frame rate is very promising up to 30 frames/second.
|