Geometric camera calibration: theory and experiments

It is known from geometric optics that any system of lenses can be approximated by a system which realizes a perspective projection of a 3D scene onto a 2D plate. Projective geometry is an efficient tool to investigate such an image forming system. This paper treats the problem of calibrating cameras using this tool. In this work, a new formulation of calibrating a moving camera is presented, and is based on the absolute conic properties. The projective invariance of the absolute conic under rigid motion ensures that its image is independent of camera position, and depends only on the intrinsic parameters of the camera. The camera calibration is computed in two steps. First, the estimation of the coefficients of the epipolar transformation is estimated. The computation of the absolute conic image is based on the pole and polar of the conic. Then, relations between intrinsic parameters and coefficients of the absolute conic image are then deduced. Both computer-generated and real data are included in experiments, which illustrates how the method works. This technique is integrated in an iterative filtering scheme