Automatic line-based estimation of radial lens distortion

Abstract

Line-based camera calibration methods estimate the radial lens distortion by using the constraint that straight lines in the 3D world must project to straight lines in the image plane, if the distortion is compensated. Thereby line-based approaches neither need a calibration pattern nor information about other camera parameters and can be applied to single views. Former line-based approaches are impaired by real curved lines in the 3D world that are misinterpreted as straight lines. This affects their robustness and reduces the accuracy of the estimation. This contribution solves the problem by a novel efficient outlier elimination of real curved lines in the 3D world. Results of studies on synthetic and natural camera images indicate a significant improvement of robustness and accuracy of the estimation by the applied outlier elimination method.