Research On Image Object Recovery And Its Application In Photograph Aircraft 3D Reconstruction

A comprehensive study was conducted in the area of "Image Object Recovery and Its Application to Aircraft 3D Reconstruction". In this dissertation image feature detection, 2D geometry extraction, object modeling from engineering drawings, camera calibration, and 3D reconstruction from photos were elaborated. In the research on 3D object reconstruction from engineering drawings, a multi-feature synthesis method was proposed to detect fuzzy corners in 2D curves, then a Corner Guided Sparse Pixel Vectorization (CGSPV) algorithm was established to extract the topology and geometry information of the drawings, after collecting 2D informations of the drawing, an iterative tree, which based on relations of 2D edges, was built to trim pathological elements, and at last, a B-Rep method was used to reconstruct object model of the drawings. Up till now, in the area of Image Based Modeling (IBM), most jobs were still confined to sequence-image-based calibration. Few contributions could be found in the camera calibration algorithm based on single view, or there indeed existed some but most of which made different kinds of assumptions to simplify the problem. In order to solve this problem, a new calibration method, which is derived from projective geometry, was proposed. This method could obtain the full parameters of the off-the-shelf camera. The calculated results of both simulated and real images were consistent with the results given by existed algorithm based on sequential images. By support of ref [87], this effective algorithm could also be applied in the unknown parallelepiped subsets, like cuboids, cube, etc. At the presupposition of known camera intrinsic parameters, the structure and metrology recovery of generic objects through a single image was still difficult. A mirror composition method was brought in this dissertation. This method was promoted by combination of the characteristic of aircraft configuration and the reconstruction theory. Theoretical analysise and experimental means were adopted to solve the relations between two intrinsic matrices corresponding to two different views. Aimed at cutting down the calculation errors introduced by image noises, appropriate numerical method was employed to meet the problem. In this section, epipolar geometry and triangulation formulae were reformulated step-by-step strictly. This method was suitable in the case of the scene-photo which is lack of vanishing points. This method can be used to calculate the point-relative-position and structure-dimension-ratio. In chapter 6 and 7, outlines of the 3D reconstruction scheme were summarized and a complete case was established. Theories and algorithms involved in chapter 2 - 5 were all impremented on PC.