Experimental Estimation of the Calibration Accuracy of Intrinsic Parameters for Digital Video Cameras

A large number of computer vision algorithms in their work use information about geometric parameters of objects in images. For such algorithms, an important element of their performance and efficiency is the calibration process. It consists of determining camera and lens parameters on the basis of which further image processing, including correction of optical distortions, is performed. This paper presents an attempt to estimate the accuracy of a standard algorithm for calibrating digital cameras by comparing its results with those of direct measurements. A series of measurements and calibrations were performed on six camera-lens pairs. The analysis of the obtained data showed that the results of calibration and direct measurements can be very different, even if the final reprojection error is small. Therefore, using calibration results with a small reprojection error does not guarantee a low measurement error due to incorrect determination of the physical parameters of the camera.