Image-Based and Landmark-Based Approach to Camera Calibration

摘要

We present two novel algorithms to calibrate a system of (an arbitrary number of) cameras each of which is described by a pinhole model augmented by a model for radial lens distortion. If a number of landmarks can be identified on images taken by these cameras, the resulting image coordinates can be measured. The task is to deduce from these measurements the locations and spatial orientations of the cameras as well as inner camera parameters (focal width, lens distortion coefficients); moreover, one is interested in assessing the accuracies of these data in terms of the (assumed) measurement accuracies and also of the accuracies of the landmark coordinates if these cannot be assumed as perfectly known. Two different approaches to accomplish this task are described, both based on differential geometric ideas to solve optimisation problems in manifolds.