Abstract
Many camera calibration methods for a non-metric digital camera have been proposed. However there is no standard procedure to evaluate an estimated distribution of image distortion directly. Calibration results are usually evaluated indirectly by such indexes as root mean square of residuals on image, three dimensional measurement errors of control points, or error estimates of obtained camera parameters. Therefore, we conducted a numerical simulation in order to examine capabilities of these indexes. In the simulation nine images were supposed to be acquired vertically to shoot control points on four layers disposed at regular intervals of the depth. Seven sets of control points were prepared, and these sets varied in the depth of the distribution of control points from 2/3 to 1/96 of the average camera height. Numerical simulation results show the limits of capabilities of these indexes. Error estimates of obtained camera parameters may be the most effective, however it is very difficult to interpret values of the error estimates obtained in a camera calibration. The authors propose that a set of numerical simulations should be conducted in a camera calibration to evaluate calibration results in addition to conventional evaluation by using the abovementioned indexes.
