Calibration And Self-calibration For The Conical Mirror Catadioptric Camera

Conical mirror catadioptric cameras have higher resolution in the periphery, lower the visual distortion and have simple shape. Conical mirror camera belongs to the non-center camera system, with respect to the central catadioptric camera model calibration is more complicated. This paper uses the non-central axis symmetric systems and approximate center model of the conical mirror studies calibration and self-calibration, to obtain the conical mirror catadioptric cameras’inner parameters and the correspondence between images, the main contents are as follows:1.For the non-central axis symmetric system of rotating symmetric mirror, finds for a given world point,there is a line in the image passing through its reflected image, direct image,the image of the mirror axis,so to utilize the invariance of cross ratio solves out the principal point, and then to utilize the principal point and image of a point solves contour circle,and then obtain the polar line of the principal point. The intersection between the polar line and the tangent of the image of the point and the intersection between the principal point and the image of the point are a pair of the vanishing point in the orthogonal direction. To use the image of two points solves the two pairs of the vanishing points in the orthogonal direction, and then solve scale factor and distortion factor of the conical mirror catadioptric camera.2. According to the principle of approximate center model. It is found that the difference between approximate model and central model is whether viewpoint and visual axis are coaxial. And conical mirror catadioptric projection has the same nature, so it can use the approximate center model and approximate epipolar geometry of conical mirror catadioptric camera to study its self-calibration. First, for the given world points takes two pictures in different location and then initialize the intrinsic parameters.And then using known intrinsic parameters estimates fundamental matrix F with eight-point algorithm with RANSAC. Finally, given the inliers,estimate the intrinsic parameters with nonlinear LM algorithm3. We performed a number of experiments with simulated and real image data using Matlab software. The experimental results show that the methods presented in this thesis is effective.