Based On The Complete Quadrangle And Using The 3D Sphere To Calibrate Paracatadioptric Camera

Camera calibration is an essential step from 2D image to 3D information in computer vision, and the basic requirement to complete visiontask. Although the field of view of common camera is limited, the central catadioptric camera can provide larger field of view, the imaging model is:the first step, a 3D point projected onto the unit sphere; The second step, the point on the unit sphere is projected onto the image plane by the virtual camera. A sphere as a common geometric object, no matter from any direction to watch it is a circle and it occluded by itself, the image of a sphere is a conic on the image plane, extracting the projection of the sphere is easier than the liner projection, it can provide high calibration precision.This thesis has studied the projection of the sphere at the central catadioptric imaging system, put forward two methods to calibrate parabolic catadioptric camera, the main research contents can be summarized as follows:1. The image of 3D sphere and the antipodal sphere image intersect four points, these four points form a complete quadrangle, two points of the four are a pair of circular points, a picture can provide two circular points,we use three pictures to calculate the camera intrinsic parameters.2. The two different points are extracted from the image of sphere,calculating their antipodal image points, antipodal image points are in the antipodal circle, these four points form a complete quadrangle and four vertices of a rectangle. Using the orthogonal properties of the rectangle form two vanishing points and determine a vanishing liner, the vanishing liner and the image of by the centre of sphere and four points as definite curve intersect at two circular points three pictures can calibrate parabolic catadioptric camera.3. Experiments with simulated and real image data are carried out using Matlab software. The catadioptric images are corrected and obtained the corresponding perspective images in the real experiment, the experimental results show the algorithm is feasible and effective.