| Summary: | 博士 === 國立成功大學 === 電腦與通信工程研究所 === 102 === This thesis explores the multiple view computer vision techniques applied to depth estimation and 3D reconstruction, including dense depth map estimation, sparse world points localization, and multiple view 3D reconstruction of piecewise planar model.
In realizing 3D display systems, an efficient and accurate estimation of disparity map between the stereo images is essential. The disparity estimation problem is commonly solved using graph cut methods, in which the disparity assignment problem is transformed to one of minimizing the global energy function. Although such an approach yields an accurate disparity map, the computational cost is relatively high. Accordingly, we propose a Hierarchical Bilateral Disparity Structure (HBDS) algorithm in which the efficiency of the GC-based method is improved without any loss in the disparity accuracy by dividing all the disparity levels hierarchically into a series of bilateral disparity structures of increasing fineness. To address the well-known ``foreground fattening' effect, a disparity refinement process is proposed comprising a fattening foreground region detection procedure followed by a disparity recovery process. The efficiency and accuracy of the proposed algorithm are verified and compared with several conventional methods using benchmark stereo images selected from the Middlebury dataset.
Due to the limitation of the conventional correspondences detection methods, locating the points on the texture-less human back using stereoscopic 3D localization technique is impracticable. To cope with the issue, the present study proposes a novel correspondences detection scheme designated as Correspondences from Epipolar geometry and Contours via Triangle barycentric coordinates (CECT). In the proposed approach, reliable correspondences are extracted from the edge contours of the human back by applying epipolar geometry and are then regarded as foundations for computing the correspondences within the edge contour based on triangle barycentric coordinates system. The accuracy and robustness of the estimated correspondences are further ensured by applying three geometric constraints. The performance of the proposed approach is demonstrated by means of a series of experiments involving 28 subjects and three different testing conditions.
An automatic 3D reconstruction method utilizing the property of inter-image homography and the concept of half-planes is proposed to produce realistic 3D model of a real-world scene portrayed in a set of images. The proposed modeling method starts with extracting the corresponding feature points and lines from the images of the world scene. Then, the extracted corresponding points and lines are filtered in accordance with region and coplanar constraints and are used to identify the correct half-planes of real-world planes. Finally, a complete 3D planar model is constructed by enlarging the half-planes to their full extent, and then merging all the extended half-planes which belong to the same world plane. The feasibility of the proposed approach is demonstrated by reconstructing 3D planar models of two real-world scenes containing objects with multiple planar facets.
|
|---|
