Intrinsic camera calibration based on radical center estimation

Camera calibration is an important step in obtaining 3D information from 2D images. Vanishing points of parallel lines have proven to be useful features for self-calibration task. Most tasks using vanishing points estimate parameters using three orthogonal vanishing points (OVPs). However, in a real scene it is hard to find views that capture a scene including three OVPs. Fortunately, in many such cases the vertical and horizontal vanishing points can still be known. Accordingly, the current paper proposes a simple, geometrically intuitive method to calibrate a camera using two orthogonal vanishing points from image streams without an assumption of the principal point is known. The Thales' theorem [16] is devised for geometric constraints and the candidate space of principal point and focal length is derived from the relation of multiple hemispheres. Through a set of experiments we demonstrate that the optimally estimated calibration can be possible.