Omnidirectional camera calibration based on an approximate factorization method

A standard perspective image can be represented with part center area of an omnidirectional image from Micusik [12], can we generalize a few camera calibration methods from the perspective image to the omnidirectional image using part information in the center area of the omnidirectional image? In this paper, a novel omnidirectional camera calibration based on the approximate factorization method from multiple images is presented. This method first performs the approximate projective factorization algorithm using part of image points, which are less distorted than other feature points, in the center area of the omnidirectional image. Then, assuming zero skew and the principal point at the center of omnidirectional image, the initial value of camera intrinsic parameters can be computed. Last, the distortion intrinsic parameters are refined on minimizing the objective function. We demonstrate this method in experiment with synthetic data and real fisheye images.