A Factorization-Based Approach to Projective Reconstruction from Line Correspondences in Multiple Images

This dissertation, "A Factorization-based Approach to Projective Reconstruction From Line Correspondences in Multiple Images" by Tuen-pui, Ng, 吳端珮, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: Abstract of thesis entitled A Factorization-Based Approach to Projective Reconstruction from Line Correspondences in Multiple Images Submitted by Ng Tuen Pui for the degree of Master of Philosophy at The University of Hong Kong in August 2004 This study addresses the problems of using line correspondences for the reconstruction of 3D scene from multiple images. Compared with planar images, a 3D model is a more informative mode of presentation. Existing methods of reconstructing 3D objects from images mainly make use of feature point correspondences across images. Straight lines, nevertheless, are more flexible in dealing with occlusion and less sensitive to noise. Besides, as most modern man- made structures are largely rectangular, straight line is an appropriate medium for describing these structures. The possibility of using straight lines as features for reconstruction is therefore a worthwhile topic for research. Before reconstruction can be carried out from an image sequence, the images must be analyzed. This process includes the detection of line features and the matching of corresponding line features across different images. As the quality of the detected and matched lines is critical to the performance of the reconstruction, this study initially explores methods of image analysis. Practical experience in edge detection was obtained in order to identify difficulties and take appropriate measures to ensure satisfactory 3D reconstruction. It was found that the main difficulty in line features extraction was to determine the end points of the lines accurately. Given the success of the factorization method in point reconstruction, an existing factorization-based method for line reconstruction was studied in detail. However, the line factorization method was found to be very sensitive to noise, and was unable to offer a practical solution to the problem. In the light of these studies, an efficient line reconstruction method is proposed. In order to take advantage of the success of factorization-based methods for points, the proposed method treats the 3D reconstruction from lines as a point-based problem in which a 3D line is determined by reconstructing two 3D points lying on it. Unlike existing factorization-based methods for points, the input points do not necessarily have to be corresponding points, as long as they lie on the corresponding lines. In other words, a measured line segment in a view can correspond to the measured line segments in other views, even though they are from different portions of a 3D line. The proposed method therefore has high tolerance for the location of end points on measured lines, and can handle occlusion in measured lines. Real and synthetic experiments have demonstrated that the proposed method converges robustly to good results, and therefore offers a feasible solution to line reconstruction. DOI: 10.5353/th_b3010949 Subjects: Image reconstruction Image processing Computer vision