Research On The Algorithms For Reconstruction Of Architecture Model Based On Image Sequences

IBM (Image based modeling) is a new technique used to recover the 3D models of real scenes from images. In this paper, a new method to recover the architectures model from image sequences is proposed. Our method is composed of the following steps: camera calibration, architecture model reconstruction based on single image and incorporation of the models in multi-scenes. Primary contributions of this thesis include: Camera calibration. According to the planar constraints, we proved that the product of the fundamental matrix and the homography matrix of an image pair is an antisymmetric matrix. Given a set of corresponding image points, which are coplanar in 3D space, a robust algorithm based on the proved antisymmetric property of the matrix is proposed. In this algorithm, the problem of calculating the homography matrix in Gang Xu's algorithm was transformed into a quadratic programming problem. Therefore, the robustness of the algorithm is improved. The geometric constraints, which are called corner structure and parallel line structure, are studied in this paper. A new algorithm for camera self-calibration is proposed by using the geometric constraints of these structures. Because only one image is used to calculate the initial solutions, the critical motion sequences can be avoided, consequentially the degeneracies in camera self-calibration can be avoided either. Then the solutions for camera calibration are optimized under the geometric constraints in two steps. Firstly, the geometric structure in one image is introduced to optimize the solutions via minimizing the differences between the observed lines and the reprojected lines. Secondly, the solutions are optimized via minimizing the differences between the observed points and the reprojected points under the geometric constraints in two images. The precisions of the calculated parameters of the cameras are improved by the optimizations in two steps. Architecture model reconstruction based on single image.Considering the regular shape of the architecture model, a new algorithm for reconstruction of architecture model based on single perspective image is proposed in this paper. The plane and the cylinder surface in real scenes are considered as therecovered primitives in this algorithm. Once the planes and cylinder surfaces are all calculated, the whole architecture model is reconstructed. The positions and normal directions of the planes in the scene are calculated one by one via interactive operation, by which the planes are recovered. The cylinder surfaces are calculated by two methods: 1. The cylinder surface is divided into several planar patches. Once all the planar patches are calculated, the cylinder surface is recovered. 2. The center and height of the cylinder are calculated, then directly to recover the cylinder. Merging the multiple partial models.In order to obtain whole architecture model, a new practical method for integrating the partial models in multi-scenes is proposed in this paper. Firstly, the list of the common coordinate frames is built with the solutions in camera calibration procedure. Then the partial models in multi-scenes are integrated with the following steps: coordinates transformation, vertices incorporation, shutting the model and plane incorporation. Through integrating the models in multi-scenes, we solve the problem of reconstructing the whole architecture model.