2008 年 13 巻 1 号 p. 7-16
The availability of software developed to correct the distortion associated with non-metric cameras was examined using a test chart composed of 63 gridiron dots and two CCD cameras: a commercially available camera as the reference and an actual surveillance camera installed in a bank. The software was designed to calculate distortion parameters such as the radial and tangential distortion coefficients and the principal points. Constant parameter values were obtained when test charts were taken at six positions in the lens field. The average of displacement errors between the 63 dots on the corrected image and those on the presumed ideal image was measured. In the reference camera, a pre-correction error of 1.24 mm decreased to 0.08 mm after correction. In the surveillance camera the error decreased from 7.92 mm to 0.33 mm. The present method was compared with two others: the graph method and the Open CV method. The former uses a distortion graph, which indicates the distortion level (%) at several locations from the center to the radial direction of the lens field. The latter is a free camera calibration software tool, developed by Intel Co. for the calculation of the intrinsic camera parameters of video camera lenses. The present method was associated with smaller error values than either of the other two methods. It is therefore suggested that this method could usefully be applied to distortion correction of facial images taken by surveillance cameras.
