Review of geometric distortion compensation in fish-eye cameras

Abstract

The majority of computer vision applications assume the pin-hole camera model. However, most optics will introduce some undesirable effects, rendering the assumption of the pin-hole camera model invalid. This is particularly evident in cameras with wide fields-of-view. The aim of distortion correction is, therefore, to transform the distorted view of wide-angle cameras to the pin-hole perspective view. Fish-eye cameras are those with "super-wide" fields-of-view, e.g. those cameras with fields-of-view of up to 180 degrees. However, these lenses exhibit severe forms of distortion. The most evident of these is radial distortion, but several other distortions, such as uneven illumination and inaccurate estimation of the centre of distortion, should also be considered when using a fish-eye camera. In this paper, we review and discuss methods of correcting radial and other distortions for fish-eye cameras and illustrate the effect of these methods on a test image exhibiting multiple types of distortion.