Stability study of camera calibration methods

Abstract

Many methods have been developed to calibrate cameras, but very few works have been done to compare such methods or to provide the user with hints on the suitability of certain algorithms under particular circumstances. This work presents a comparative analysis of eight methods of calibration for cameras using a pattern as reference. This paper concentrates on the stability and accuracy of these methods when the pattern is relocated or the camera con guration varies. This study was carried out with real and synthetic images and using, whenever possible, the code made available by the methods' authors on the WWW. The experiments demonstrate that most of these methods are not stable in the sense that the intrinsic parameters returned by a calibration method su ered important variations under small displacements of the camera relative to the calibration pattern. Similar results were obtained for extrinsic parameters when the camera only changed its internal con guration (i.e. when it zooms in or out) but kept constant its relative position to the calibration pattern. In addition, this study shows that the image disparity is not an indicator of the reliability of the methods. In spite of the fact that the majority of the methods showed similar levels of global error, the calibrated values obtained for intrinsic and extrinsic parameters varied substantially among these methods for the same set of calibration images.