Panoramic Depth Reconstruction within a Single Shot byOptimizing Global Sphere Radii

• Main Authors: Po-Yao Huang, 黃柏耀
• Other Authors: Ming Ouhyang
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/z23nff

碩士 === 國立臺灣大學 === 資訊工程學研究所 === 106 === In the world of virtual reality, depth estimation scene reconstruction is always one of the most important issue of VR. Our experiment is about depth map scene reconstruction with a low cost camera facilities, but with a different matching cost map reconstruction. We use our experiment to compare with the other depth estimation matching cost method from the paper a low cost portable polycamera for stereoscopic 360 imaging. The two methods implement similar principle such as camera calibration and spherical rectification, but in the depth estimation part, we use the other local method to define the depth value and labelling so as to calculate the matching cost map. The other method uses trinocular matching cost map reconstruction, which define the angle between the point X to the two neighboring cameras as disparity. That’s different from our method which define the spherical radius as disparity. We implement the two kinds of experiment and compare the two matching cost values and see the different between them. The purpose of our experiment is to find a best way to minimize the depth estimation matching cost and generate the panoramic scene images with a low cost facilities. We implement the method on the paper ”A Low-Cost Portable Polycamera for Stereoscopic 360 Imaging” and found some problem of it. So I changed the disparity and use another algorithm with local method to modify the defect of the method.