英文摘要 | Computer vision research is focused on recovering the structure and motion of 3D world from the 2D image of scene to go to perform some guided vision tasks. The theoretical main steps include camera calibration, motion analysis and 3D reconstruction, where camera calibration is the most basic and essential part. In many computer vision applications, such as visual surveillance, virtual reality and robot navigation, central catadioptric camera with quite a large field of view, is widely used. Therefore more and more attentions are paid to the central catadioptric camera calibration. However, due to the large distortions, there are still many difficulties on catadioptric camera calibration. In this thesis, we study the geometry of central catadioptric images and then propose some central catadioptric camera calibration methods. The following is our main work: (1) We prove that the central catadioptric projection of a circle is a quartic curve. Afterward, according to a circle with different size or location, we find that this quartic curve can also be reduced to the image of a point, line or sphere. The reduced result, being equivalent to the previous existing equations, and the circle imaging provides a unified imaging theory of different geometric elements. Then, under this unified imaging theory, we discuss the central catadioptric camera calibration based on different geometric elements. (2) We propose a new catadioptric line fitting method, which is suitable to all kinds of central catadioptric cameras and has higher accuracy than state of the art methods. Due to the partial occlusion,it is very difficult to correctly estimate the central catadioptric line image from its visible part. What's more, the existing method in the literatures cannot still solve this problem effectively. In addition, we find that if the antipodal points of image points on the visible arc are known, the fitting accuracy can be improved greatly. Based on this idea, we propose a new method for fitting line images, which can be applied to all central catadioptric projection. Firstly, a new relationship between a pair of antipodal image points and the camera principal point are derived. Next,using this relationship, a new method is proposed to estimate the line images. Finally, the camera intrinsic parameters are calibrated using the estimated conics. We also find under paracatadioptric camera, this method additionally with the necessary and sufficient conditions ... |
修改评论