| In this thesis, we first discuss the camera self-calibration method based on the correspondence of image points. The solution is found to be very sensitive to image noise; Furthermore, since the camera self-calibration is carried out solely on the basis of image correspondence, outliers due to point mismatches can completely spoil the consequent estimation of the unknown parameters.; In view of these problems, our research focuses on improving the robustness with respect to image noise and outliers. To this end, we first introduce a non-linear approach to the camera self-calibration using points. We then apply a robust least-median of squares (LMS) method to handle outliers. Finally, we introduce a new camera self-calibration algorithm using ellipses. Compared to points, ellipses can basically eliminate the point mismatch and image occlusion problems. The performance of these algorithms is validated using the synthetic and real image data. |
