Learning priors for calibrating families of stereo cameras

Online camera recalibration is necessary for long-term deploymentof computervision systems. Existing algorithms assume that the source of recalibration information is a set of features in a general 3D scene; and that enough features are observed that the calibration problem is well- constrained. However, these assumptions are frequently in- valid outside the laboratory. Real-world scenes often lack texture, contain repeated texture, or are mostly planar, mak- ing calibration difficult or impossible. In this paper we consider the calibration of families of stereo cameras, where each camera is assumed to have pa- rameters drawn from a common but unknown prior dis- tribution. We show how estimation of this prior using a small-number of offline-calibrated cameras (e.g. from the same production line) allows online calibration of addi- tional cameras using a small number of point correspon- dences; and that using the estimated prior significantly in- creases the accuracy and robustness of stereo camera cali- bration.