Omnidirectional cameras are widely used in robot navigation, virtual reality, surveillance, image-based rendering, image-based metrology and 3D reconstruction thanks to their very large field of view. Since the camera calibration is an indispensable step to recover 3D information from 2D images and is necessary for effective image distortion correction, the thesis is focused on omnidirectional camera calibration. The main work can be summarized as follows: 1. A new fisheye lenses distortion correction method using spherical perspective projection (SPP) constraint is proposed. Firstly, polynomial distortion model with free parameters are selected; Secondly, these parameters are estimated based on the fact that the projected image curves of space lines should be mapped to great circles on a unit spherical surface, i.e., the SPP constraint. 2. A novel method for calibration of central catadioptric cameras using geometric invariants is proposed. We prove that the projection of a line can provide three invariants whereas the projection of a sphere can provide two. From these invariants, constraint equations for the intrinsic parameters of catadioptric camera are derived. 3. A unified imaging model for central catadioptric cameras and fisheye camera is proposed. An important merit of our new unified model is that existing calibration methods for central catadioptric cameras can be directly applied to fisheye cameras. 4. We prove that with a calibrated catadioptric camera, the projection of a space line is a conic, and this conic had only two free parameters, and the projected conic of a space sphere has 3 free parameters. Based on these results, an effective Hough transform based method is proposed to extract these conics. 5. In recent panorama-based virtual reality systems, a key issue is how to achieve continuous walkthrough between the two neighboring panoramas. A new approach is proposed to deal with the problem as: Along the transition path, sampling images are taken at first, then the images in the non-sampling points are synthesized from those sampling images using an image transformation method. A prototype system of walkthrough in large environment based on chained panoramas is implemented based on our proposed approach, and the results are satisfactory.
修改评论