Accurate camera calibration with distortion models using sphere images

In order to improve the accuracy of sphere images based camera calibration, a novel approach was proposed in this paper which can calibrate both linear parameters and distortion coefficients simultaneously. The great axis and bitangent lines of the projection conics are applied to solve principal point and sphere center projections. Then the focal length is computed using rotational symmetry of projective cone. Finally distortion coefficients are estimated by optimization search algorithm. Synthetic data experiments analyzed the main error factors and real data results showed that the re-projection error of the proposed method was less than 0.1 pixel which seems to be more accurate than existing methods using spheres and reach the same accuracy level as the planar target method.