Research Of The Method For Configuration And Calibration Of Parabolic Camera

With the rapid development of technology, a camera with a large view field is more and more desirable than a normal camera with limited visual field. From now on, there are mainly two major methods to realize wide view-field imaging, one is the method of real-time omni-directional imaging and another is the method of multiple images mosaic. The catadioptric imaging model studied in this paper belongs to the former. After studying and comparing the reported articles, we choose the parabolic mirror as reflective mirror in the catadioptric imaging system. To simplify the imaging model and enable to conveniently transform its image into a perspective image through mathematical model, we make the system satisfying both the constraint of central projection and the constraint of single viewpoint. For the parabolic camera system, with the known intrinsic parameters, we propose a calibration method that is fast and efficient for the remainder parameters.Precisely speaking, two major problems are explored in this thesis:(1) How to efficiently configure a parabolic camera composed by one parabolic mirror and one traditional CCD camera, which satisfies the constraint of single viewpoint and central projection.(2) Based on the configured parabolic camera, how to design a fast and efficient calibration algorithm for the remainder parameters with the known internal parameters of the camera.Firstly, the relationship between various configuration parameters and its corresponding system errors is tested and depicted using numerical simulation experiments, from which the range for best configuration parameters is obtained. Based on the configured parabolic camera, we propose a rapid calibration algorithm for the parabolic camera with the known intrinsic parameters of the camera. The core idea of the calibration algorithm is that the calibration problem is formulated into a polynomial eigenvalue problem via the plane-induced Homographic matrix.Then an analytic solution is obtained to avoid the complicated numerical computation and the possible divergent case, and hence rapid calibration algorithm is achieved. The results also show that five or more pairs of corresponding image points are sufficient for solving this problem. Finally, numerical simulation and real data experiments are performed to validate the accuracy and robustness of the proposed algorithm and analyzed from the quantitative and qualitative viewpoints.The conclusion is that satisfactory results are obtained in those experiments, which indicate that our algorithm can be applied in various real scenes, such as visual surveillance, the Simultaneous Localization and Mapping (SLAM) and so on.