• Journal of the Korean Society for Precision Engineering
• /
• v.16 no.2 s.95
• /
• pp.164-175
• /
• 1999

If the camera plane is nearly parallel to the calibration board on which objects are defined, most of existing calibration approaches such as Tsai's radial-alignment-constraint method cannot be applied. Recently, for such an ill-conditioned case, Zhuang & Wu suggested the linear two-stage calibration algorithm assuming that the exact values of focal length and scale factor are known a priori. In this paper, we developed an iterative two-stage algorithm starts with initial guess fo the two parameters to determine the value of the others using Zhuang & Wu's method. In the second stage, the two parameters are locally optimized. This process is repeated until any improvement cannot be expected any more. The performance comparison between Zhuang & Wu's method and our algorithm shows the superiority of ours. Also included are the computational results for the effects of the distribution and the number of calibration points on the calibration performance.

• Journal of Institute of Control, Robotics and Systems
• /
• v.7 no.12
• /
• pp.1001-1008
• /
• 2001

This paper presents a new camera calibration algorithm for ill-conditioned cases in which the camera plane is nearly parallel to a set of non-coplanar calibration boards. for the ill-conditioned case, most of existing calibration approaches such as Tsais radial-alignment-constraint method cannot be applied. Recently, for the ill-conditioned coplanar calibration Lee&Lee[16] proposed an iterative algorithm based on the least square method. The non-coplanar calibration algorithm presented in this paper is an iterative two-stage procedure with extends the previous coplanar calibration algorithm. Through the first stage, camera, position and orientation parameters as well as one radial distortion factor are determined optimally for a given data of the scale factor and the focal length. In the second stage, the scale factor and the focal length are locally optimized. This process is repeated until any improvement cannot be expected any more Computational results are provided to show the performance of the algorithm developed.