An Image-Based Virtual Reality Based on Panoramic Stereo Imaging

• Main Authors: Ding Li-Chune, 丁力群
• Other Authors: Chou Chun-Hsien
• Format: Others
• Language: zh-TW
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/50124290077488226940

碩士 === 大同工學院 === 電機工程研究所 === 86 === With the speedy progress in computational power of personal computers and the fast growing interest in 3-D multimedia, the technologies of generating, transmitting and displaying virtual reality (VR) have recently received intensive attention. The ongoing development will supply users with the fine human-computer interface that can generate the natural and interactive virtual environment. Although the most VRs are constructed by building the 3D models for all objects, this approach has some disadvantages. The cost including the time to build the models and theequipment is expensive. The reality of the virtual world also is not satisfied by the participants, especially for nature objects, including trees, flowers and organism etc. Because of these reasons, an image-based VR system is proposed in this thesis. Through the proposed system, the virtual environment is quite real because it is built from the natural images. An important property of the image-based VR is that no matter how the complexity of the scenery, it would not have any effect on virtual world.In this thesis, the panoramic images are generated by arranging natural photographs taken at a number of view points around a rotational axis. In order to have the stereo vision, this system uses two cameras to capture the left-right image pair for each sampled view. However, these captured images can not be stitched directly due to the optical and geometrical distortion. The methods usedto solve these problems include: detection of overlappingareas, image warping, luminance and chrominance blending,and manual postprocessing. Overlapping areas of the images to be stitched would supply important information for later automatic stitching and blending. Image warping reduces the distortion causedby image plane of different orientations. After image warping, the images would be stitched automatically. Thetwo postprocessing steps are applied in this system. Oneis alignment of the corresponding pixels which are not registered and the another is finely tuning the warping to each of four quadrants of the stitched image. Finally,the blending would smooth the difference of the colorsbetween the images to be stitched. Simulation results show that good quality stereoscopic panoramic images. The crystal eye also would be appliedto display the panoramic images. Through it, the stereovirtual world would be shown completely.