A novel calibration method with iterative correction

Abstract

The implicit calibration is always needed for many visual hull reconstruction in a canonical world coordinate. We use a translational planar object for camera calibration instead of the object with known 3D geometry. However, there is an unavoidable inaccuracy in manufacture and assembly of planar object, that leads to the planar object deviate from the space of 3D calibration object. The deviation of planar object causes an error of control point's world coordinate. This error will decrease calibration precision directly. We propose a correction approach by which the accurate camera calibration is achieved. The original information of control point is used to compute an initial set of camera parameters. With these parameters, we calculate a skew-correction coefficient matrix by which the error of information of the control point is corrected. The new information is used to optimize a set of new camera parameters. This process is then repeated until convergence. In many experiments based on real images, the pixel reprojection errors obtained by our method are about 30% lower than those of traditional method. Increased accuracy of camera calibration directly leads to improvement in 3D reconstruction.