Stratified Solving Of Metric Reconstruction

3D reconstruction is the final step of simulation of the human eye functions in computer vision, in other words recover 3D information of the objects. It is an important and active research area. It has been widely used in industrial inspection, military, medical, aerospace, entertainment and other fields. In the case of camera intrinsic parameters remain unchanged; there are two major categories of 3D Reconstruction framework. First, the intrinsic can be calculated based on the fundamental matrix and epipolar geometry. Usually the Kruppa can be established between two images, then, camera intrinsic parameters are solved; Second, the major method is stratified and gradually self-calibration reconstruction, that,do projective reconstruction of the image sequence are first, on this basis, it can update to affine reconstruction and European reconstruction .According to the general framework of 3D reconstruction, this paper major research is the problem of solving the fundamental matrix, solving the infinite homograph matrix and the reconstruction result under two kinds of framework. Through analyzing the references, it can found that: the traditional 8-point algorithm does not guarantee the fundamental matrix rank is 2; solving infinite homograph need to solve the vanishing point and vanishing line; the solving result of intrinsic under two kinds of framework are similar or consistent.The main contributions of this paper are as follows:1. The algorithm of linear solving the fundamental matrix has been elevated. On the basis of advantage and disadvantage of the linear algorithm and nonlinear algorithm, a linear solution of the fundamental matrix is rank 2 provided. At first, our algorithm use the thoughts of the improved 8-Points algorithm which normalizes the coordinates of the image matching points, and combines with characteristics of the fundamental matrix rank 2, then gets an equivalent model of the rank 2 fundamental matrix. At last, the equation and the fundamental matrix model is gotten by analyzing the geometric relationship of the epipole constraints. The linear algorithm solves the parameters and the epipole coordinates. The experimental has been analyzed and compared.2. The algorithm of calculating the infinite homograph matrix has been elevated. Analyze the epipolar geometry can get a equation which is a new constraint between the homograph of the plane at infinity and the epipolar point. From the new constraint we can solve the infinite homograph matrix with linear method. Then we can get affine reconstruction through triangulation principle. As compared to the literature, this method avoid solving the vanish points and vanish lines, and do not need the information of the homograph of a space plane. If we know the correspondence poles and points of the two images, we can solve the infinite homograph matrix. Then we can use the relationship between absolute conic and infinite homograph matrix to solve the camera's intrinsic parameters. The feasibility of the method can be seen from the simulation experiments; Real experiments show that the method has really high robustness.3. The results of reconstruction under two kinds of framework has been compared. In the same experimental conditions, the comparison of the intrinsic parameters which solved by two kinds of framework show: the intrinsic parameters are similarity.