When Should We Consider Lens Distortion in Camera Calibration

In this paper, calibrating a camera refers to the process of determining the transformation between a 3D object coordinate system and the 2D image coordinate system of the camera. Knowing this transformation, 3D position of a feature point seen in two cameras can be determined by using triangulation. Suppose a distortion-free lens is had, this 3D-2D transformation can be well approximated by a linear model. Unfortunately, most off-the-shelf lenses have a certain amount of distortion. If our goal is to get highly accurate 3D measurements, we have to considcr lcns distortion in camcra calibration. However, considering lens distortion will not only complicate the camcra calibration procedure, but also complicate the subsequent on-line processing such as feature-point comspondence and camera re-calibration. It is hence desirable to avoid considering lens distortion whenever the distortion-induced error is tolerable. This work investigates the effect of neglecting lens distortion, and proposes a simple efficient calibration algorithm which can either exclude or include lens distortion -depending on whether one can neglect lens distortion and still meet the accuracy requirement.