Research On Camera Calibration Of 3D Information Extraction

Camera calibration is a significant field of computer vision and it has been widely applied in many fields, such as robot vision navigation, military application, mechanical manufacture, virtual reality and so on. By the theory of Marr vision, research on camera calibration is emphasized. A system of 3D information extraction based on binocular stereo vision is designed successfully to extract 3D coordinates of feature points. The main job of the thesis can be described as follows:A new approach for camera calibration based on vanishing points is proposed. Two pairs of vanishing points of orthogonal directions from each image are obtained by using the concept of cross ratio and harmonic conjugate in projective geometry and the orthogonal property of the two diagonals on a square. All intrinsic parameters are calculated linearly after capturing the template from at least three shooting angles. Errors of edge detection and line fitting are avoided because the method does not require image matching. Thus the proposed method can reduce the calculation complexity and improve the precision of calibration greatly.An improved method for camera calibration based on circular points is also proposed. The vanishing points on four diametric directions are determined via Harris corners on the template images. The image coordinates of circular points are calculated by the special relation between vanishing points and circular points. All the intrinsic parameters can be calculated accurately by constraints of circular points and intrinsic parameter matrix. The advantages of this method are not only no extraction of the ellipse and four diameters, but also no complicated calculation of circular points via the ellipse and the vanishing lines.Based on feature points matching between a pair of stereo images, one of the camera calibration methods proposed above is then applied into the system of 3D information extraction. 3D coordinates of feature points are obtained successfully. There are few errors between 3D coordinates of feature points obtained and their standard values. Experimental results have shown that the method proposed in the paper is robust and the system of 3D information extraction has a good result.