Robust camera calibration using 2D to 3D feature correspondences

The problem of calibrating a vision system is extremely important for practical applications such as 3D reconstruction and pose estimation of three-dimensional objects. In this paper we present a method for optimally estimating the internal and external camera parameters from point or line correspondences. First we extend the linear method of Faugeras & Toscani (1986) to line correspondences, second, we develop for both point and line correspondences a new method based on the minimization of an error function. We show how to minimize this error function in order to compute the unknown parameters (i.e., the internal and external camera parameters). This minimization leads to a non-linear optimization problem. We introduce an elegant way to automatically establish 2D to 3D feature correspondences using projective geometry. We perform a stability analysis both for our method and for the linear method. In the light Of this comparison, the non-linear method seems to be the most robust one with respect to noise. Finally, we present some experiments to completely calibrate a binocular stereo rig with different calibration patterns. This stereo rig forms the main vision sensor of the Janus humanoid robotics system we are currently developing.