The question of accuracy with geometric camera calibration

In the field of machine vision, camera calibration refers to the experimental determination of a set of parameters which describe the image formation process for a given analytical model of the machine vision system. An accurate, reliable calibration procedure is essential for most industrial machine vision applications including mechanical metrology, robot assembly, reverse engineering, stereo vision etc. One of the most systematic calibration procedures for 3D machine vision applications was proposed by Heikkila in which a comprehensive set of camera parameters is automatically evaluated by observing a calibration target consisting of two perpendicular planes, each with 256 circular control points. Other similar techniques employ a checkerboard pattern as a target and use the vertices of the squares as control points. While these techniques are sound from a theoretical point of view, they do not adequately speak to the question of measurement accuracy. The objective of this work is to gain and understanding of the problems associated with Geometric Camera Calibration through the application of Design of Experiments. A response surface methodology, namely a CCD Design, is carried to analyze the effects. This paper also highlights the issue of calibration accuracy by addressing the following fundamental question: Assuming a certain tolerance or uncertainty in the calibration target, what is the expected error with respect to the measured camera parameters and what is the impact on the final 3D machine vision application?