Active-Camera Calibration Using Weakly-Localized Image Features

This work presents a new approach to the problem of calibrating a zooming camera. Calibration information is necessary for active cameras to perform tasks that require known camera parameters such as stereo reconstruction. Our method for calibrating a zoom camera solves for the parameters of a camera model using a staged global optimization technique. The input is a sequence of images of a known calibration target obtained for diierent mechanical zoom settings. Our new approach addressed two central problems that arise in applying classical calibration algorithms directly to the multiple-image, active camera domain. First, the process is designed to avoid heavy dependence on individual, strongly localized features. Feature localization is instead included as part of the error measure used in various passes of the optimization process. Second, images are not calibrated independently. Rather, the staged optimization process considers all 1 images simultaneously, representing the parameters of the nal calibrated camera as a function of zoom. This calibration method is completely automated, requiring only basic assumptions about the position and structure of the calibration pattern and the read-back of the mechanical zoom lens settings. In this paper we describe the details of the approach and show experimental results from a complete implementation using both synthetic and real data. Our results on real images from an active, zooming stereo rig demonstrate that with this method one can localize epipolar correspondence at any zoom setting to an accuracy of less than 1 pixel. Abstract This work presents a new approach to the problem of calibrating a zooming camera. Calibration information is necessary for active cameras to perform tasks that require known camera parameters such as stereo reconstruction. Our method for calibrating a zoom camera solves for the parameters of a camera model using a staged global optimization technique. The input is a sequence of images of a known calibration target obtained for diierent mechanical zoom settings. Our new approach addressed two central problems that arise in applying classical calibration algorithms directly to the multiple-image, active camera domain. First, the process is designed to avoid heavy dependence on individual, strongly localized features. Feature localization is instead included as part of the error measure used in various passes of the optimization process. Second, images are not calibrated independently. Rather, the staged optimization process considers all images simultaneously, representing the parameters of the nal calibrated camera as a function of zoom. This calibration method is completely …